fetch_impl.hh revision 7849:2290428b5f04 
1/*
2 * Copyright (c) 2010 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 * All rights reserved.
16 *
17 * Redistribution and use in source and binary forms, with or without
18 * modification, are permitted provided that the following conditions are
19 * met: redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer;
21 * redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution;
24 * neither the name of the copyright holders nor the names of its
25 * contributors may be used to endorse or promote products derived from
26 * this software without specific prior written permission.
27 *
28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 *
40 * Authors: Kevin Lim
41 *          Korey Sewell
42 */
43
44#include <algorithm>
45#include <cstring>
46
47#include "arch/isa_traits.hh"
48#include "arch/utility.hh"
49#include "base/types.hh"
50#include "config/the_isa.hh"
51#include "config/use_checker.hh"
52#include "cpu/checker/cpu.hh"
53#include "cpu/exetrace.hh"
54#include "cpu/o3/fetch.hh"
55#include "mem/packet.hh"
56#include "mem/request.hh"
57#include "params/DerivO3CPU.hh"
58#include "sim/byteswap.hh"
59#include "sim/core.hh"
60
61#if FULL_SYSTEM
62#include "arch/tlb.hh"
63#include "arch/vtophys.hh"
64#include "sim/system.hh"
65#endif // FULL_SYSTEM
66
67using namespace std;
68
69template<class Impl>
70void
71DefaultFetch<Impl>::IcachePort::setPeer(Port *port)
72{
73    Port::setPeer(port);
74
75    fetch->setIcache();
76}
77
78template<class Impl>
79Tick
80DefaultFetch<Impl>::IcachePort::recvAtomic(PacketPtr pkt)
81{
82    panic("DefaultFetch doesn't expect recvAtomic callback!");
83    return curTick();
84}
85
86template<class Impl>
87void
88DefaultFetch<Impl>::IcachePort::recvFunctional(PacketPtr pkt)
89{
90    DPRINTF(Fetch, "DefaultFetch doesn't update its state from a "
91            "functional call.");
92}
93
94template<class Impl>
95void
96DefaultFetch<Impl>::IcachePort::recvStatusChange(Status status)
97{
98    if (status == RangeChange) {
99        if (!snoopRangeSent) {
100            snoopRangeSent = true;
101            sendStatusChange(Port::RangeChange);
102        }
103        return;
104    }
105
106    panic("DefaultFetch doesn't expect recvStatusChange callback!");
107}
108
109template<class Impl>
110bool
111DefaultFetch<Impl>::IcachePort::recvTiming(PacketPtr pkt)
112{
113    DPRINTF(Fetch, "Received timing\n");
114    if (pkt->isResponse()) {
115        fetch->processCacheCompletion(pkt);
116    }
117    //else Snooped a coherence request, just return
118    return true;
119}
120
121template<class Impl>
122void
123DefaultFetch<Impl>::IcachePort::recvRetry()
124{
125    fetch->recvRetry();
126}
127
128template<class Impl>
129DefaultFetch<Impl>::DefaultFetch(O3CPU *_cpu, DerivO3CPUParams *params)
130    : cpu(_cpu),
131      branchPred(params),
132      predecoder(NULL),
133      decodeToFetchDelay(params->decodeToFetchDelay),
134      renameToFetchDelay(params->renameToFetchDelay),
135      iewToFetchDelay(params->iewToFetchDelay),
136      commitToFetchDelay(params->commitToFetchDelay),
137      fetchWidth(params->fetchWidth),
138      cacheBlocked(false),
139      retryPkt(NULL),
140      retryTid(InvalidThreadID),
141      numThreads(params->numThreads),
142      numFetchingThreads(params->smtNumFetchingThreads),
143      interruptPending(false),
144      drainPending(false),
145      switchedOut(false)
146{
147    if (numThreads > Impl::MaxThreads)
148        fatal("numThreads (%d) is larger than compiled limit (%d),\n"
149              "\tincrease MaxThreads in src/cpu/o3/impl.hh\n",
150              numThreads, static_cast<int>(Impl::MaxThreads));
151
152    // Set fetch stage's status to inactive.
153    _status = Inactive;
154
155    std::string policy = params->smtFetchPolicy;
156
157    // Convert string to lowercase
158    std::transform(policy.begin(), policy.end(), policy.begin(),
159                   (int(*)(int)) tolower);
160
161    // Figure out fetch policy
162    if (policy == "singlethread") {
163        fetchPolicy = SingleThread;
164        if (numThreads > 1)
165            panic("Invalid Fetch Policy for a SMT workload.");
166    } else if (policy == "roundrobin") {
167        fetchPolicy = RoundRobin;
168        DPRINTF(Fetch, "Fetch policy set to Round Robin\n");
169    } else if (policy == "branch") {
170        fetchPolicy = Branch;
171        DPRINTF(Fetch, "Fetch policy set to Branch Count\n");
172    } else if (policy == "iqcount") {
173        fetchPolicy = IQ;
174        DPRINTF(Fetch, "Fetch policy set to IQ count\n");
175    } else if (policy == "lsqcount") {
176        fetchPolicy = LSQ;
177        DPRINTF(Fetch, "Fetch policy set to LSQ count\n");
178    } else {
179        fatal("Invalid Fetch Policy. Options Are: {SingleThread,"
180              " RoundRobin,LSQcount,IQcount}\n");
181    }
182
183    // Get the size of an instruction.
184    instSize = sizeof(TheISA::MachInst);
185
186    // Name is finally available, so create the port.
187    icachePort = new IcachePort(this);
188
189    icachePort->snoopRangeSent = false;
190
191#if USE_CHECKER
192    if (cpu->checker) {
193        cpu->checker->setIcachePort(icachePort);
194    }
195#endif
196}
197
198template <class Impl>
199std::string
200DefaultFetch<Impl>::name() const
201{
202    return cpu->name() + ".fetch";
203}
204
205template <class Impl>
206void
207DefaultFetch<Impl>::regStats()
208{
209    icacheStallCycles
210        .name(name() + ".icacheStallCycles")
211        .desc("Number of cycles fetch is stalled on an Icache miss")
212        .prereq(icacheStallCycles);
213
214    fetchedInsts
215        .name(name() + ".Insts")
216        .desc("Number of instructions fetch has processed")
217        .prereq(fetchedInsts);
218
219    fetchedBranches
220        .name(name() + ".Branches")
221        .desc("Number of branches that fetch encountered")
222        .prereq(fetchedBranches);
223
224    predictedBranches
225        .name(name() + ".predictedBranches")
226        .desc("Number of branches that fetch has predicted taken")
227        .prereq(predictedBranches);
228
229    fetchCycles
230        .name(name() + ".Cycles")
231        .desc("Number of cycles fetch has run and was not squashing or"
232              " blocked")
233        .prereq(fetchCycles);
234
235    fetchSquashCycles
236        .name(name() + ".SquashCycles")
237        .desc("Number of cycles fetch has spent squashing")
238        .prereq(fetchSquashCycles);
239
240    fetchTlbCycles
241        .name(name() + ".TlbCycles")
242        .desc("Number of cycles fetch has spent waiting for tlb")
243        .prereq(fetchTlbCycles);
244
245    fetchIdleCycles
246        .name(name() + ".IdleCycles")
247        .desc("Number of cycles fetch was idle")
248        .prereq(fetchIdleCycles);
249
250    fetchBlockedCycles
251        .name(name() + ".BlockedCycles")
252        .desc("Number of cycles fetch has spent blocked")
253        .prereq(fetchBlockedCycles);
254
255    fetchedCacheLines
256        .name(name() + ".CacheLines")
257        .desc("Number of cache lines fetched")
258        .prereq(fetchedCacheLines);
259
260    fetchMiscStallCycles
261        .name(name() + ".MiscStallCycles")
262        .desc("Number of cycles fetch has spent waiting on interrupts, or "
263              "bad addresses, or out of MSHRs")
264        .prereq(fetchMiscStallCycles);
265
266    fetchIcacheSquashes
267        .name(name() + ".IcacheSquashes")
268        .desc("Number of outstanding Icache misses that were squashed")
269        .prereq(fetchIcacheSquashes);
270
271    fetchNisnDist
272        .init(/* base value */ 0,
273              /* last value */ fetchWidth,
274              /* bucket size */ 1)
275        .name(name() + ".rateDist")
276        .desc("Number of instructions fetched each cycle (Total)")
277        .flags(Stats::pdf);
278
279    idleRate
280        .name(name() + ".idleRate")
281        .desc("Percent of cycles fetch was idle")
282        .prereq(idleRate);
283    idleRate = fetchIdleCycles * 100 / cpu->numCycles;
284
285    branchRate
286        .name(name() + ".branchRate")
287        .desc("Number of branch fetches per cycle")
288        .flags(Stats::total);
289    branchRate = fetchedBranches / cpu->numCycles;
290
291    fetchRate
292        .name(name() + ".rate")
293        .desc("Number of inst fetches per cycle")
294        .flags(Stats::total);
295    fetchRate = fetchedInsts / cpu->numCycles;
296
297    branchPred.regStats();
298}
299
300template<class Impl>
301void
302DefaultFetch<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *time_buffer)
303{
304    timeBuffer = time_buffer;
305
306    // Create wires to get information from proper places in time buffer.
307    fromDecode = timeBuffer->getWire(-decodeToFetchDelay);
308    fromRename = timeBuffer->getWire(-renameToFetchDelay);
309    fromIEW = timeBuffer->getWire(-iewToFetchDelay);
310    fromCommit = timeBuffer->getWire(-commitToFetchDelay);
311}
312
313template<class Impl>
314void
315DefaultFetch<Impl>::setActiveThreads(std::list<ThreadID> *at_ptr)
316{
317    activeThreads = at_ptr;
318}
319
320template<class Impl>
321void
322DefaultFetch<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr)
323{
324    fetchQueue = fq_ptr;
325
326    // Create wire to write information to proper place in fetch queue.
327    toDecode = fetchQueue->getWire(0);
328}
329
330template<class Impl>
331void
332DefaultFetch<Impl>::initStage()
333{
334    // Setup PC and nextPC with initial state.
335    for (ThreadID tid = 0; tid < numThreads; tid++) {
336        pc[tid] = cpu->pcState(tid);
337        fetchOffset[tid] = 0;
338        macroop[tid] = NULL;
339    }
340
341    for (ThreadID tid = 0; tid < numThreads; tid++) {
342
343        fetchStatus[tid] = Running;
344
345        priorityList.push_back(tid);
346
347        memReq[tid] = NULL;
348
349        stalls[tid].decode = false;
350        stalls[tid].rename = false;
351        stalls[tid].iew = false;
352        stalls[tid].commit = false;
353    }
354
355    // Schedule fetch to get the correct PC from the CPU
356    // scheduleFetchStartupEvent(1);
357
358    // Fetch needs to start fetching instructions at the very beginning,
359    // so it must start up in active state.
360    switchToActive();
361}
362
363template<class Impl>
364void
365DefaultFetch<Impl>::setIcache()
366{
367    // Size of cache block.
368    cacheBlkSize = icachePort->peerBlockSize();
369
370    // Create mask to get rid of offset bits.
371    cacheBlkMask = (cacheBlkSize - 1);
372
373    for (ThreadID tid = 0; tid < numThreads; tid++) {
374        // Create space to store a cache line.
375        cacheData[tid] = new uint8_t[cacheBlkSize];
376        cacheDataPC[tid] = 0;
377        cacheDataValid[tid] = false;
378    }
379}
380
381template<class Impl>
382void
383DefaultFetch<Impl>::processCacheCompletion(PacketPtr pkt)
384{
385    ThreadID tid = pkt->req->threadId();
386
387    DPRINTF(Fetch, "[tid:%u] Waking up from cache miss.\n",tid);
388
389    assert(!pkt->wasNacked());
390
391    // Only change the status if it's still waiting on the icache access
392    // to return.
393    if (fetchStatus[tid] != IcacheWaitResponse ||
394        pkt->req != memReq[tid] ||
395        isSwitchedOut()) {
396        ++fetchIcacheSquashes;
397        delete pkt->req;
398        delete pkt;
399        return;
400    }
401
402    memcpy(cacheData[tid], pkt->getPtr<uint8_t>(), cacheBlkSize);
403    cacheDataValid[tid] = true;
404
405    if (!drainPending) {
406        // Wake up the CPU (if it went to sleep and was waiting on
407        // this completion event).
408        cpu->wakeCPU();
409
410        DPRINTF(Activity, "[tid:%u] Activating fetch due to cache completion\n",
411                tid);
412
413        switchToActive();
414    }
415
416    // Only switch to IcacheAccessComplete if we're not stalled as well.
417    if (checkStall(tid)) {
418        fetchStatus[tid] = Blocked;
419    } else {
420        fetchStatus[tid] = IcacheAccessComplete;
421    }
422
423    // Reset the mem req to NULL.
424    delete pkt->req;
425    delete pkt;
426    memReq[tid] = NULL;
427}
428
429template <class Impl>
430bool
431DefaultFetch<Impl>::drain()
432{
433    // Fetch is ready to drain at any time.
434    cpu->signalDrained();
435    drainPending = true;
436    return true;
437}
438
439template <class Impl>
440void
441DefaultFetch<Impl>::resume()
442{
443    drainPending = false;
444}
445
446template <class Impl>
447void
448DefaultFetch<Impl>::switchOut()
449{
450    switchedOut = true;
451    // Branch predictor needs to have its state cleared.
452    branchPred.switchOut();
453}
454
455template <class Impl>
456void
457DefaultFetch<Impl>::takeOverFrom()
458{
459    // Reset all state
460    for (ThreadID i = 0; i < Impl::MaxThreads; ++i) {
461        stalls[i].decode = 0;
462        stalls[i].rename = 0;
463        stalls[i].iew = 0;
464        stalls[i].commit = 0;
465        pc[i] = cpu->pcState(i);
466        fetchStatus[i] = Running;
467    }
468    numInst = 0;
469    wroteToTimeBuffer = false;
470    _status = Inactive;
471    switchedOut = false;
472    interruptPending = false;
473    branchPred.takeOverFrom();
474}
475
476template <class Impl>
477void
478DefaultFetch<Impl>::wakeFromQuiesce()
479{
480    DPRINTF(Fetch, "Waking up from quiesce\n");
481    // Hopefully this is safe
482    // @todo: Allow other threads to wake from quiesce.
483    fetchStatus[0] = Running;
484}
485
486template <class Impl>
487inline void
488DefaultFetch<Impl>::switchToActive()
489{
490    if (_status == Inactive) {
491        DPRINTF(Activity, "Activating stage.\n");
492
493        cpu->activateStage(O3CPU::FetchIdx);
494
495        _status = Active;
496    }
497}
498
499template <class Impl>
500inline void
501DefaultFetch<Impl>::switchToInactive()
502{
503    if (_status == Active) {
504        DPRINTF(Activity, "Deactivating stage.\n");
505
506        cpu->deactivateStage(O3CPU::FetchIdx);
507
508        _status = Inactive;
509    }
510}
511
512template <class Impl>
513bool
514DefaultFetch<Impl>::lookupAndUpdateNextPC(
515        DynInstPtr &inst, TheISA::PCState &nextPC)
516{
517    // Do branch prediction check here.
518    // A bit of a misnomer...next_PC is actually the current PC until
519    // this function updates it.
520    bool predict_taken;
521
522    if (!inst->isControl()) {
523        TheISA::advancePC(nextPC, inst->staticInst);
524        inst->setPredTarg(nextPC);
525        inst->setPredTaken(false);
526        return false;
527    }
528
529    ThreadID tid = inst->threadNumber;
530    predict_taken = branchPred.predict(inst, nextPC, tid);
531
532    if (predict_taken) {
533        DPRINTF(Fetch, "[tid:%i]: [sn:%i]:  Branch predicted to be taken to %s.\n",
534                tid, inst->seqNum, nextPC);
535    } else {
536        DPRINTF(Fetch, "[tid:%i]: [sn:%i]:Branch predicted to be not taken.\n",
537                tid, inst->seqNum);
538    }
539
540    DPRINTF(Fetch, "[tid:%i]: [sn:%i] Branch predicted to go to %s.\n",
541            tid, inst->seqNum, nextPC);
542    inst->setPredTarg(nextPC);
543    inst->setPredTaken(predict_taken);
544
545    ++fetchedBranches;
546
547    if (predict_taken) {
548        ++predictedBranches;
549    }
550
551    return predict_taken;
552}
553
554template <class Impl>
555bool
556DefaultFetch<Impl>::fetchCacheLine(Addr vaddr, ThreadID tid, Addr pc)
557{
558    Fault fault = NoFault;
559
560    // @todo: not sure if these should block translation.
561    //AlphaDep
562    if (cacheBlocked) {
563        DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, cache blocked\n",
564                tid);
565        return false;
566    } else if (isSwitchedOut()) {
567        DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, switched out\n",
568                tid);
569        return false;
570    } else if (checkInterrupt(pc)) {
571        // Hold off fetch from getting new instructions when:
572        // Cache is blocked, or
573        // while an interrupt is pending and we're not in PAL mode, or
574        // fetch is switched out.
575        DPRINTF(Fetch, "[tid:%i] Can't fetch cache line, interrupt pending\n",
576                tid);
577        return false;
578    }
579
580    // Align the fetch address so it's at the start of a cache block.
581    Addr block_PC = icacheBlockAlignPC(vaddr);
582
583    // Setup the memReq to do a read of the first instruction's address.
584    // Set the appropriate read size and flags as well.
585    // Build request here.
586    RequestPtr mem_req =
587        new Request(tid, block_PC, cacheBlkSize, Request::INST_FETCH,
588                    pc, cpu->thread[tid]->contextId(), tid);
589
590    memReq[tid] = mem_req;
591
592    // Initiate translation of the icache block
593    fetchStatus[tid] = ItlbWait;
594    FetchTranslation *trans = new FetchTranslation(this);
595    cpu->itb->translateTiming(mem_req, cpu->thread[tid]->getTC(),
596                              trans, BaseTLB::Execute);
597    return true;
598}
599
600template <class Impl>
601void
602DefaultFetch<Impl>::finishTranslation(Fault fault, RequestPtr mem_req)
603{
604    ThreadID tid = mem_req->threadId();
605    Addr block_PC = mem_req->getVaddr();
606
607    // If translation was successful, attempt to read the icache block.
608    if (fault == NoFault) {
609        // Build packet here.
610        PacketPtr data_pkt = new Packet(mem_req,
611                                        MemCmd::ReadReq, Packet::Broadcast);
612        data_pkt->dataDynamicArray(new uint8_t[cacheBlkSize]);
613
614        cacheDataPC[tid] = block_PC;
615        cacheDataValid[tid] = false;
616        DPRINTF(Fetch, "Fetch: Doing instruction read.\n");
617
618        fetchedCacheLines++;
619
620        // Access the cache.
621        if (!icachePort->sendTiming(data_pkt)) {
622            assert(retryPkt == NULL);
623            assert(retryTid == InvalidThreadID);
624            DPRINTF(Fetch, "[tid:%i] Out of MSHRs!\n", tid);
625
626            fetchStatus[tid] = IcacheWaitRetry;
627            retryPkt = data_pkt;
628            retryTid = tid;
629            cacheBlocked = true;
630        } else {
631            DPRINTF(Fetch, "[tid:%i]: Doing Icache access.\n", tid);
632            DPRINTF(Activity, "[tid:%i]: Activity: Waiting on I-cache "
633                    "response.\n", tid);
634
635            lastIcacheStall[tid] = curTick();
636            fetchStatus[tid] = IcacheWaitResponse;
637        }
638    } else {
639        // Translation faulted, icache request won't be sent.
640        delete mem_req;
641        memReq[tid] = NULL;
642
643        // Send the fault to commit.  This thread will not do anything
644        // until commit handles the fault.  The only other way it can
645        // wake up is if a squash comes along and changes the PC.
646        TheISA::PCState fetchPC = pc[tid];
647
648        // We will use a nop in ordier to carry the fault.
649        DynInstPtr instruction = buildInst(tid,
650                StaticInstPtr(TheISA::NoopMachInst, fetchPC.instAddr()),
651                NULL, fetchPC, fetchPC, false);
652
653        instruction->setPredTarg(fetchPC);
654        instruction->fault = fault;
655        wroteToTimeBuffer = true;
656
657        fetchStatus[tid] = TrapPending;
658
659        DPRINTF(Fetch, "[tid:%i]: Blocked, need to handle the trap.\n", tid);
660        DPRINTF(Fetch, "[tid:%i]: fault (%s) detected @ PC %s.\n",
661                tid, fault->name(), pc[tid]);
662    }
663    _status = updateFetchStatus();
664}
665
666template <class Impl>
667inline void
668DefaultFetch<Impl>::doSquash(const TheISA::PCState &newPC, ThreadID tid)
669{
670    DPRINTF(Fetch, "[tid:%i]: Squashing, setting PC to: %s.\n",
671            tid, newPC);
672
673    pc[tid] = newPC;
674    fetchOffset[tid] = 0;
675    macroop[tid] = NULL;
676    predecoder.reset();
677
678    // Clear the icache miss if it's outstanding.
679    if (fetchStatus[tid] == IcacheWaitResponse) {
680        DPRINTF(Fetch, "[tid:%i]: Squashing outstanding Icache miss.\n",
681                tid);
682        memReq[tid] = NULL;
683    }
684
685    // Get rid of the retrying packet if it was from this thread.
686    if (retryTid == tid) {
687        assert(cacheBlocked);
688        if (retryPkt) {
689            delete retryPkt->req;
690            delete retryPkt;
691        }
692        retryPkt = NULL;
693        retryTid = InvalidThreadID;
694    }
695
696    fetchStatus[tid] = Squashing;
697
698    ++fetchSquashCycles;
699}
700
701template<class Impl>
702void
703DefaultFetch<Impl>::squashFromDecode(const TheISA::PCState &newPC,
704                                     const InstSeqNum &seq_num, ThreadID tid)
705{
706    DPRINTF(Fetch, "[tid:%i]: Squashing from decode.\n", tid);
707
708    doSquash(newPC, tid);
709
710    // Tell the CPU to remove any instructions that are in flight between
711    // fetch and decode.
712    cpu->removeInstsUntil(seq_num, tid);
713}
714
715template<class Impl>
716bool
717DefaultFetch<Impl>::checkStall(ThreadID tid) const
718{
719    bool ret_val = false;
720
721    if (cpu->contextSwitch) {
722        DPRINTF(Fetch,"[tid:%i]: Stalling for a context switch.\n",tid);
723        ret_val = true;
724    } else if (stalls[tid].decode) {
725        DPRINTF(Fetch,"[tid:%i]: Stall from Decode stage detected.\n",tid);
726        ret_val = true;
727    } else if (stalls[tid].rename) {
728        DPRINTF(Fetch,"[tid:%i]: Stall from Rename stage detected.\n",tid);
729        ret_val = true;
730    } else if (stalls[tid].iew) {
731        DPRINTF(Fetch,"[tid:%i]: Stall from IEW stage detected.\n",tid);
732        ret_val = true;
733    } else if (stalls[tid].commit) {
734        DPRINTF(Fetch,"[tid:%i]: Stall from Commit stage detected.\n",tid);
735        ret_val = true;
736    }
737
738    return ret_val;
739}
740
741template<class Impl>
742typename DefaultFetch<Impl>::FetchStatus
743DefaultFetch<Impl>::updateFetchStatus()
744{
745    //Check Running
746    list<ThreadID>::iterator threads = activeThreads->begin();
747    list<ThreadID>::iterator end = activeThreads->end();
748
749    while (threads != end) {
750        ThreadID tid = *threads++;
751
752        if (fetchStatus[tid] == Running ||
753            fetchStatus[tid] == Squashing ||
754            fetchStatus[tid] == IcacheAccessComplete) {
755
756            if (_status == Inactive) {
757                DPRINTF(Activity, "[tid:%i]: Activating stage.\n",tid);
758
759                if (fetchStatus[tid] == IcacheAccessComplete) {
760                    DPRINTF(Activity, "[tid:%i]: Activating fetch due to cache"
761                            "completion\n",tid);
762                }
763
764                cpu->activateStage(O3CPU::FetchIdx);
765            }
766
767            return Active;
768        }
769    }
770
771    // Stage is switching from active to inactive, notify CPU of it.
772    if (_status == Active) {
773        DPRINTF(Activity, "Deactivating stage.\n");
774
775        cpu->deactivateStage(O3CPU::FetchIdx);
776    }
777
778    return Inactive;
779}
780
781template <class Impl>
782void
783DefaultFetch<Impl>::squash(const TheISA::PCState &newPC,
784                           const InstSeqNum &seq_num, ThreadID tid)
785{
786    DPRINTF(Fetch, "[tid:%u]: Squash from commit.\n", tid);
787
788    doSquash(newPC, tid);
789
790    // Tell the CPU to remove any instructions that are not in the ROB.
791    cpu->removeInstsNotInROB(tid);
792}
793
794template <class Impl>
795void
796DefaultFetch<Impl>::tick()
797{
798    list<ThreadID>::iterator threads = activeThreads->begin();
799    list<ThreadID>::iterator end = activeThreads->end();
800    bool status_change = false;
801
802    wroteToTimeBuffer = false;
803
804    while (threads != end) {
805        ThreadID tid = *threads++;
806
807        // Check the signals for each thread to determine the proper status
808        // for each thread.
809        bool updated_status = checkSignalsAndUpdate(tid);
810        status_change =  status_change || updated_status;
811    }
812
813    DPRINTF(Fetch, "Running stage.\n");
814
815    // Reset the number of the instruction we're fetching.
816    numInst = 0;
817
818#if FULL_SYSTEM
819    if (fromCommit->commitInfo[0].interruptPending) {
820        interruptPending = true;
821    }
822
823    if (fromCommit->commitInfo[0].clearInterrupt) {
824        interruptPending = false;
825    }
826#endif
827
828    for (threadFetched = 0; threadFetched < numFetchingThreads;
829         threadFetched++) {
830        // Fetch each of the actively fetching threads.
831        fetch(status_change);
832    }
833
834    // Record number of instructions fetched this cycle for distribution.
835    fetchNisnDist.sample(numInst);
836
837    if (status_change) {
838        // Change the fetch stage status if there was a status change.
839        _status = updateFetchStatus();
840    }
841
842    // If there was activity this cycle, inform the CPU of it.
843    if (wroteToTimeBuffer || cpu->contextSwitch) {
844        DPRINTF(Activity, "Activity this cycle.\n");
845
846        cpu->activityThisCycle();
847    }
848}
849
850template <class Impl>
851bool
852DefaultFetch<Impl>::checkSignalsAndUpdate(ThreadID tid)
853{
854    // Update the per thread stall statuses.
855    if (fromDecode->decodeBlock[tid]) {
856        stalls[tid].decode = true;
857    }
858
859    if (fromDecode->decodeUnblock[tid]) {
860        assert(stalls[tid].decode);
861        assert(!fromDecode->decodeBlock[tid]);
862        stalls[tid].decode = false;
863    }
864
865    if (fromRename->renameBlock[tid]) {
866        stalls[tid].rename = true;
867    }
868
869    if (fromRename->renameUnblock[tid]) {
870        assert(stalls[tid].rename);
871        assert(!fromRename->renameBlock[tid]);
872        stalls[tid].rename = false;
873    }
874
875    if (fromIEW->iewBlock[tid]) {
876        stalls[tid].iew = true;
877    }
878
879    if (fromIEW->iewUnblock[tid]) {
880        assert(stalls[tid].iew);
881        assert(!fromIEW->iewBlock[tid]);
882        stalls[tid].iew = false;
883    }
884
885    if (fromCommit->commitBlock[tid]) {
886        stalls[tid].commit = true;
887    }
888
889    if (fromCommit->commitUnblock[tid]) {
890        assert(stalls[tid].commit);
891        assert(!fromCommit->commitBlock[tid]);
892        stalls[tid].commit = false;
893    }
894
895    // Check squash signals from commit.
896    if (fromCommit->commitInfo[tid].squash) {
897
898        DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash "
899                "from commit.\n",tid);
900        // In any case, squash.
901        squash(fromCommit->commitInfo[tid].pc,
902               fromCommit->commitInfo[tid].doneSeqNum,
903               tid);
904
905        // Also check if there's a mispredict that happened.
906        if (fromCommit->commitInfo[tid].branchMispredict) {
907            branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum,
908                              fromCommit->commitInfo[tid].pc,
909                              fromCommit->commitInfo[tid].branchTaken,
910                              tid);
911        } else {
912            branchPred.squash(fromCommit->commitInfo[tid].doneSeqNum,
913                              tid);
914        }
915
916        return true;
917    } else if (fromCommit->commitInfo[tid].doneSeqNum) {
918        // Update the branch predictor if it wasn't a squashed instruction
919        // that was broadcasted.
920        branchPred.update(fromCommit->commitInfo[tid].doneSeqNum, tid);
921    }
922
923    // Check ROB squash signals from commit.
924    if (fromCommit->commitInfo[tid].robSquashing) {
925        DPRINTF(Fetch, "[tid:%u]: ROB is still squashing.\n", tid);
926
927        // Continue to squash.
928        fetchStatus[tid] = Squashing;
929
930        return true;
931    }
932
933    // Check squash signals from decode.
934    if (fromDecode->decodeInfo[tid].squash) {
935        DPRINTF(Fetch, "[tid:%u]: Squashing instructions due to squash "
936                "from decode.\n",tid);
937
938        // Update the branch predictor.
939        if (fromDecode->decodeInfo[tid].branchMispredict) {
940            branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum,
941                              fromDecode->decodeInfo[tid].nextPC,
942                              fromDecode->decodeInfo[tid].branchTaken,
943                              tid);
944        } else {
945            branchPred.squash(fromDecode->decodeInfo[tid].doneSeqNum,
946                              tid);
947        }
948
949        if (fetchStatus[tid] != Squashing) {
950
951            TheISA::PCState nextPC = fromDecode->decodeInfo[tid].nextPC;
952            DPRINTF(Fetch, "Squashing from decode with PC = %s\n", nextPC);
953            // Squash unless we're already squashing
954            squashFromDecode(fromDecode->decodeInfo[tid].nextPC,
955                             fromDecode->decodeInfo[tid].doneSeqNum,
956                             tid);
957
958            return true;
959        }
960    }
961
962    if (checkStall(tid) &&
963        fetchStatus[tid] != IcacheWaitResponse &&
964        fetchStatus[tid] != IcacheWaitRetry) {
965        DPRINTF(Fetch, "[tid:%i]: Setting to blocked\n",tid);
966
967        fetchStatus[tid] = Blocked;
968
969        return true;
970    }
971
972    if (fetchStatus[tid] == Blocked ||
973        fetchStatus[tid] == Squashing) {
974        // Switch status to running if fetch isn't being told to block or
975        // squash this cycle.
976        DPRINTF(Fetch, "[tid:%i]: Done squashing, switching to running.\n",
977                tid);
978
979        fetchStatus[tid] = Running;
980
981        return true;
982    }
983
984    // If we've reached this point, we have not gotten any signals that
985    // cause fetch to change its status.  Fetch remains the same as before.
986    return false;
987}
988
989template<class Impl>
990typename Impl::DynInstPtr
991DefaultFetch<Impl>::buildInst(ThreadID tid, StaticInstPtr staticInst,
992                              StaticInstPtr curMacroop, TheISA::PCState thisPC,
993                              TheISA::PCState nextPC, bool trace)
994{
995    // Get a sequence number.
996    InstSeqNum seq = cpu->getAndIncrementInstSeq();
997
998    // Create a new DynInst from the instruction fetched.
999    DynInstPtr instruction =
1000        new DynInst(staticInst, thisPC, nextPC, seq, cpu);
1001    instruction->setTid(tid);
1002
1003    instruction->setASID(tid);
1004
1005    instruction->setThreadState(cpu->thread[tid]);
1006
1007    DPRINTF(Fetch, "[tid:%i]: Instruction PC %#x (%d) created "
1008            "[sn:%lli]\n", tid, thisPC.instAddr(),
1009            thisPC.microPC(), seq);
1010
1011    DPRINTF(Fetch, "[tid:%i]: Instruction is: %s\n", tid,
1012            instruction->staticInst->
1013            disassemble(thisPC.instAddr()));
1014
1015#if TRACING_ON
1016    if (trace) {
1017        instruction->traceData =
1018            cpu->getTracer()->getInstRecord(curTick(), cpu->tcBase(tid),
1019                    instruction->staticInst, thisPC, curMacroop);
1020    }
1021#else
1022    instruction->traceData = NULL;
1023#endif
1024
1025    // Add instruction to the CPU's list of instructions.
1026    instruction->setInstListIt(cpu->addInst(instruction));
1027
1028    // Write the instruction to the first slot in the queue
1029    // that heads to decode.
1030    assert(numInst < fetchWidth);
1031    toDecode->insts[toDecode->size++] = instruction;
1032
1033    return instruction;
1034}
1035
1036template<class Impl>
1037void
1038DefaultFetch<Impl>::fetch(bool &status_change)
1039{
1040    //////////////////////////////////////////
1041    // Start actual fetch
1042    //////////////////////////////////////////
1043    ThreadID tid = getFetchingThread(fetchPolicy);
1044
1045    if (tid == InvalidThreadID || drainPending) {
1046        DPRINTF(Fetch,"There are no more threads available to fetch from.\n");
1047
1048        // Breaks looping condition in tick()
1049        threadFetched = numFetchingThreads;
1050        return;
1051    }
1052
1053    DPRINTF(Fetch, "Attempting to fetch from [tid:%i]\n", tid);
1054
1055    // The current PC.
1056    TheISA::PCState thisPC = pc[tid];
1057
1058    Addr pcOffset = fetchOffset[tid];
1059    Addr fetchAddr = (thisPC.instAddr() + pcOffset) & BaseCPU::PCMask;
1060
1061    // If returning from the delay of a cache miss, then update the status
1062    // to running, otherwise do the cache access.  Possibly move this up
1063    // to tick() function.
1064    if (fetchStatus[tid] == IcacheAccessComplete) {
1065        DPRINTF(Fetch, "[tid:%i]: Icache miss is complete.\n", tid);
1066
1067        fetchStatus[tid] = Running;
1068        status_change = true;
1069    } else if (fetchStatus[tid] == Running) {
1070        // Align the fetch PC so its at the start of a cache block.
1071        Addr block_PC = icacheBlockAlignPC(fetchAddr);
1072
1073        // Unless buffer already got the block, fetch it from icache.
1074        if (!cacheDataValid[tid] || block_PC != cacheDataPC[tid]) {
1075            DPRINTF(Fetch, "[tid:%i]: Attempting to translate and read "
1076                    "instruction, starting at PC %s.\n", tid, thisPC);
1077
1078            fetchCacheLine(fetchAddr, tid, thisPC.instAddr());
1079
1080            if (fetchStatus[tid] == IcacheWaitResponse)
1081                ++icacheStallCycles;
1082            else if (fetchStatus[tid] == ItlbWait)
1083                ++fetchTlbCycles;
1084            else
1085                ++fetchMiscStallCycles;
1086            return;
1087        } else if (checkInterrupt(thisPC.instAddr()) || isSwitchedOut()) {
1088            ++fetchMiscStallCycles;
1089            return;
1090        }
1091    } else {
1092        if (fetchStatus[tid] == Idle) {
1093            ++fetchIdleCycles;
1094            DPRINTF(Fetch, "[tid:%i]: Fetch is idle!\n", tid);
1095        } else if (fetchStatus[tid] == Blocked) {
1096            ++fetchBlockedCycles;
1097            DPRINTF(Fetch, "[tid:%i]: Fetch is blocked!\n", tid);
1098        } else if (fetchStatus[tid] == Squashing) {
1099            ++fetchSquashCycles;
1100            DPRINTF(Fetch, "[tid:%i]: Fetch is squashing!\n", tid);
1101        } else if (fetchStatus[tid] == IcacheWaitResponse) {
1102            ++icacheStallCycles;
1103            DPRINTF(Fetch, "[tid:%i]: Fetch is waiting cache response!\n",
1104                    tid);
1105        } else if (fetchStatus[tid] == ItlbWait) {
1106            DPRINTF(Fetch, "[tid:%i]: Fetch is waiting ITLB walk to "
1107                    "finish! \n", tid);
1108            ++fetchTlbCycles;
1109        }
1110
1111        // Status is Idle, Squashing, Blocked, ItlbWait or IcacheWaitResponse
1112        // so fetch should do nothing.
1113        return;
1114    }
1115
1116    ++fetchCycles;
1117
1118    TheISA::PCState nextPC = thisPC;
1119
1120    StaticInstPtr staticInst = NULL;
1121    StaticInstPtr curMacroop = macroop[tid];
1122
1123    // If the read of the first instruction was successful, then grab the
1124    // instructions from the rest of the cache line and put them into the
1125    // queue heading to decode.
1126
1127    DPRINTF(Fetch, "[tid:%i]: Adding instructions to queue to "
1128            "decode.\n", tid);
1129
1130    // Need to keep track of whether or not a predicted branch
1131    // ended this fetch block.
1132    bool predictedBranch = false;
1133
1134    TheISA::MachInst *cacheInsts =
1135        reinterpret_cast<TheISA::MachInst *>(cacheData[tid]);
1136
1137    const unsigned numInsts = cacheBlkSize / instSize;
1138    unsigned blkOffset = (fetchAddr - cacheDataPC[tid]) / instSize;
1139
1140    // Loop through instruction memory from the cache.
1141    while (blkOffset < numInsts &&
1142           numInst < fetchWidth &&
1143           !predictedBranch) {
1144
1145        // If we need to process more memory, do it now.
1146        if (!curMacroop && !predecoder.extMachInstReady()) {
1147            if (ISA_HAS_DELAY_SLOT && pcOffset == 0) {
1148                // Walk past any annulled delay slot instructions.
1149                Addr pcAddr = thisPC.instAddr() & BaseCPU::PCMask;
1150                while (fetchAddr != pcAddr && blkOffset < numInsts) {
1151                    blkOffset++;
1152                    fetchAddr += instSize;
1153                }
1154                if (blkOffset >= numInsts)
1155                    break;
1156            }
1157            MachInst inst = TheISA::gtoh(cacheInsts[blkOffset]);
1158
1159            predecoder.setTC(cpu->thread[tid]->getTC());
1160            predecoder.moreBytes(thisPC, fetchAddr, inst);
1161
1162            if (predecoder.needMoreBytes()) {
1163                blkOffset++;
1164                fetchAddr += instSize;
1165                pcOffset += instSize;
1166            }
1167        }
1168
1169        // Extract as many instructions and/or microops as we can from
1170        // the memory we've processed so far.
1171        do {
1172            if (!curMacroop) {
1173                if (predecoder.extMachInstReady()) {
1174                    ExtMachInst extMachInst;
1175
1176                    extMachInst = predecoder.getExtMachInst(thisPC);
1177                    pcOffset = 0;
1178                    staticInst = StaticInstPtr(extMachInst,
1179                                               thisPC.instAddr());
1180
1181                    // Increment stat of fetched instructions.
1182                    ++fetchedInsts;
1183
1184                    if (staticInst->isMacroop())
1185                        curMacroop = staticInst;
1186                } else {
1187                    // We need more bytes for this instruction.
1188                    break;
1189                }
1190            }
1191            if (curMacroop) {
1192                staticInst = curMacroop->fetchMicroop(thisPC.microPC());
1193                if (staticInst->isLastMicroop())
1194                    curMacroop = NULL;
1195            }
1196
1197            DynInstPtr instruction =
1198                buildInst(tid, staticInst, curMacroop,
1199                          thisPC, nextPC, true);
1200
1201            numInst++;
1202
1203            nextPC = thisPC;
1204
1205            // If we're branching after this instruction, quite fetching
1206            // from the same block then.
1207            predictedBranch |= thisPC.branching();
1208            predictedBranch |=
1209                lookupAndUpdateNextPC(instruction, nextPC);
1210            if (predictedBranch) {
1211                DPRINTF(Fetch, "Branch detected with PC = %s\n", thisPC);
1212            }
1213
1214            // Move to the next instruction, unless we have a branch.
1215            thisPC = nextPC;
1216
1217            if (instruction->isQuiesce()) {
1218                DPRINTF(Fetch,
1219                        "Quiesce instruction encountered, halting fetch!");
1220                fetchStatus[tid] = QuiescePending;
1221                status_change = true;
1222                break;
1223            }
1224        } while ((curMacroop || predecoder.extMachInstReady()) &&
1225                 numInst < fetchWidth);
1226    }
1227
1228    if (predictedBranch) {
1229        DPRINTF(Fetch, "[tid:%i]: Done fetching, predicted branch "
1230                "instruction encountered.\n", tid);
1231    } else if (numInst >= fetchWidth) {
1232        DPRINTF(Fetch, "[tid:%i]: Done fetching, reached fetch bandwidth "
1233                "for this cycle.\n", tid);
1234    } else if (blkOffset >= cacheBlkSize) {
1235        DPRINTF(Fetch, "[tid:%i]: Done fetching, reached the end of cache "
1236                "block.\n", tid);
1237    }
1238
1239    macroop[tid] = curMacroop;
1240    fetchOffset[tid] = pcOffset;
1241
1242    if (numInst > 0) {
1243        wroteToTimeBuffer = true;
1244    }
1245
1246    pc[tid] = thisPC;
1247}
1248
1249template<class Impl>
1250void
1251DefaultFetch<Impl>::recvRetry()
1252{
1253    if (retryPkt != NULL) {
1254        assert(cacheBlocked);
1255        assert(retryTid != InvalidThreadID);
1256        assert(fetchStatus[retryTid] == IcacheWaitRetry);
1257
1258        if (icachePort->sendTiming(retryPkt)) {
1259            fetchStatus[retryTid] = IcacheWaitResponse;
1260            retryPkt = NULL;
1261            retryTid = InvalidThreadID;
1262            cacheBlocked = false;
1263        }
1264    } else {
1265        assert(retryTid == InvalidThreadID);
1266        // Access has been squashed since it was sent out.  Just clear
1267        // the cache being blocked.
1268        cacheBlocked = false;
1269    }
1270}
1271
1272///////////////////////////////////////
1273//                                   //
1274//  SMT FETCH POLICY MAINTAINED HERE //
1275//                                   //
1276///////////////////////////////////////
1277template<class Impl>
1278ThreadID
1279DefaultFetch<Impl>::getFetchingThread(FetchPriority &fetch_priority)
1280{
1281    if (numThreads > 1) {
1282        switch (fetch_priority) {
1283
1284          case SingleThread:
1285            return 0;
1286
1287          case RoundRobin:
1288            return roundRobin();
1289
1290          case IQ:
1291            return iqCount();
1292
1293          case LSQ:
1294            return lsqCount();
1295
1296          case Branch:
1297            return branchCount();
1298
1299          default:
1300            return InvalidThreadID;
1301        }
1302    } else {
1303        list<ThreadID>::iterator thread = activeThreads->begin();
1304        if (thread == activeThreads->end()) {
1305            return InvalidThreadID;
1306        }
1307
1308        ThreadID tid = *thread;
1309
1310        if (fetchStatus[tid] == Running ||
1311            fetchStatus[tid] == IcacheAccessComplete ||
1312            fetchStatus[tid] == Idle) {
1313            return tid;
1314        } else {
1315            return InvalidThreadID;
1316        }
1317    }
1318}
1319
1320
1321template<class Impl>
1322ThreadID
1323DefaultFetch<Impl>::roundRobin()
1324{
1325    list<ThreadID>::iterator pri_iter = priorityList.begin();
1326    list<ThreadID>::iterator end      = priorityList.end();
1327
1328    ThreadID high_pri;
1329
1330    while (pri_iter != end) {
1331        high_pri = *pri_iter;
1332
1333        assert(high_pri <= numThreads);
1334
1335        if (fetchStatus[high_pri] == Running ||
1336            fetchStatus[high_pri] == IcacheAccessComplete ||
1337            fetchStatus[high_pri] == Idle) {
1338
1339            priorityList.erase(pri_iter);
1340            priorityList.push_back(high_pri);
1341
1342            return high_pri;
1343        }
1344
1345        pri_iter++;
1346    }
1347
1348    return InvalidThreadID;
1349}
1350
1351template<class Impl>
1352ThreadID
1353DefaultFetch<Impl>::iqCount()
1354{
1355    std::priority_queue<ThreadID> PQ;
1356
1357    list<ThreadID>::iterator threads = activeThreads->begin();
1358    list<ThreadID>::iterator end = activeThreads->end();
1359
1360    while (threads != end) {
1361        ThreadID tid = *threads++;
1362
1363        PQ.push(fromIEW->iewInfo[tid].iqCount);
1364    }
1365
1366    while (!PQ.empty()) {
1367        ThreadID high_pri = PQ.top();
1368
1369        if (fetchStatus[high_pri] == Running ||
1370            fetchStatus[high_pri] == IcacheAccessComplete ||
1371            fetchStatus[high_pri] == Idle)
1372            return high_pri;
1373        else
1374            PQ.pop();
1375
1376    }
1377
1378    return InvalidThreadID;
1379}
1380
1381template<class Impl>
1382ThreadID
1383DefaultFetch<Impl>::lsqCount()
1384{
1385    std::priority_queue<ThreadID> PQ;
1386
1387    list<ThreadID>::iterator threads = activeThreads->begin();
1388    list<ThreadID>::iterator end = activeThreads->end();
1389
1390    while (threads != end) {
1391        ThreadID tid = *threads++;
1392
1393        PQ.push(fromIEW->iewInfo[tid].ldstqCount);
1394    }
1395
1396    while (!PQ.empty()) {
1397        ThreadID high_pri = PQ.top();
1398
1399        if (fetchStatus[high_pri] == Running ||
1400            fetchStatus[high_pri] == IcacheAccessComplete ||
1401            fetchStatus[high_pri] == Idle)
1402            return high_pri;
1403        else
1404            PQ.pop();
1405    }
1406
1407    return InvalidThreadID;
1408}
1409
1410template<class Impl>
1411ThreadID
1412DefaultFetch<Impl>::branchCount()
1413{
1414#if 0
1415    list<ThreadID>::iterator thread = activeThreads->begin();
1416    assert(thread != activeThreads->end());
1417    ThreadID tid = *thread;
1418#endif
1419
1420    panic("Branch Count Fetch policy unimplemented\n");
1421    return InvalidThreadID;
1422}
1423