lsq_unit_impl.hh revision 4326
1/*
2 * Copyright (c) 2004-2005 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 "arch/locked_mem.hh"
33#include "config/use_checker.hh"
34
35#include "cpu/o3/lsq.hh"
36#include "cpu/o3/lsq_unit.hh"
37#include "base/str.hh"
38#include "mem/packet.hh"
39#include "mem/request.hh"
40
41#if USE_CHECKER
42#include "cpu/checker/cpu.hh"
43#endif
44
45template<class Impl>
46LSQUnit<Impl>::WritebackEvent::WritebackEvent(DynInstPtr &_inst, PacketPtr _pkt,
47                                              LSQUnit *lsq_ptr)
48    : Event(&mainEventQueue), inst(_inst), pkt(_pkt), lsqPtr(lsq_ptr)
49{
50    this->setFlags(Event::AutoDelete);
51}
52
53template<class Impl>
54void
55LSQUnit<Impl>::WritebackEvent::process()
56{
57    if (!lsqPtr->isSwitchedOut()) {
58        lsqPtr->writeback(inst, pkt);
59    }
60    delete pkt;
61}
62
63template<class Impl>
64const char *
65LSQUnit<Impl>::WritebackEvent::description()
66{
67    return "Store writeback event";
68}
69
70template<class Impl>
71void
72LSQUnit<Impl>::completeDataAccess(PacketPtr pkt)
73{
74    LSQSenderState *state = dynamic_cast<LSQSenderState *>(pkt->senderState);
75    DynInstPtr inst = state->inst;
76    DPRINTF(IEW, "Writeback event [sn:%lli]\n", inst->seqNum);
77    DPRINTF(Activity, "Activity: Writeback event [sn:%lli]\n", inst->seqNum);
78
79    //iewStage->ldstQueue.removeMSHR(inst->threadNumber,inst->seqNum);
80
81    if (isSwitchedOut() || inst->isSquashed()) {
82        iewStage->decrWb(inst->seqNum);
83        delete state;
84        delete pkt->req;
85        delete pkt;
86        return;
87    } else {
88        if (!state->noWB) {
89            writeback(inst, pkt);
90        }
91
92        if (inst->isStore()) {
93            completeStore(state->idx);
94        }
95    }
96
97    delete state;
98    delete pkt->req;
99    delete pkt;
100}
101
102template <class Impl>
103LSQUnit<Impl>::LSQUnit()
104    : loads(0), stores(0), storesToWB(0), stalled(false),
105      isStoreBlocked(false), isLoadBlocked(false),
106      loadBlockedHandled(false)
107{
108}
109
110template<class Impl>
111void
112LSQUnit<Impl>::init(Params *params, LSQ *lsq_ptr, unsigned maxLQEntries,
113                    unsigned maxSQEntries, unsigned id)
114{
115//    DPRINTF(LSQUnit, "Creating LSQUnit%i object.\n",id);
116
117    switchedOut = false;
118
119    lsq = lsq_ptr;
120
121    lsqID = id;
122
123    // Add 1 for the sentinel entry (they are circular queues).
124    LQEntries = maxLQEntries + 1;
125    SQEntries = maxSQEntries + 1;
126
127    loadQueue.resize(LQEntries);
128    storeQueue.resize(SQEntries);
129
130    loadHead = loadTail = 0;
131
132    storeHead = storeWBIdx = storeTail = 0;
133
134    usedPorts = 0;
135    cachePorts = params->cachePorts;
136
137    retryPkt = NULL;
138    memDepViolator = NULL;
139
140    blockedLoadSeqNum = 0;
141}
142
143template<class Impl>
144void
145LSQUnit<Impl>::setCPU(O3CPU *cpu_ptr)
146{
147    cpu = cpu_ptr;
148
149#if USE_CHECKER
150    if (cpu->checker) {
151        cpu->checker->setDcachePort(dcachePort);
152    }
153#endif
154}
155
156template<class Impl>
157std::string
158LSQUnit<Impl>::name() const
159{
160    if (Impl::MaxThreads == 1) {
161        return iewStage->name() + ".lsq";
162    } else {
163        return iewStage->name() + ".lsq.thread." + to_string(lsqID);
164    }
165}
166
167template<class Impl>
168void
169LSQUnit<Impl>::regStats()
170{
171    lsqForwLoads
172        .name(name() + ".forwLoads")
173        .desc("Number of loads that had data forwarded from stores");
174
175    invAddrLoads
176        .name(name() + ".invAddrLoads")
177        .desc("Number of loads ignored due to an invalid address");
178
179    lsqSquashedLoads
180        .name(name() + ".squashedLoads")
181        .desc("Number of loads squashed");
182
183    lsqIgnoredResponses
184        .name(name() + ".ignoredResponses")
185        .desc("Number of memory responses ignored because the instruction is squashed");
186
187    lsqMemOrderViolation
188        .name(name() + ".memOrderViolation")
189        .desc("Number of memory ordering violations");
190
191    lsqSquashedStores
192        .name(name() + ".squashedStores")
193        .desc("Number of stores squashed");
194
195    invAddrSwpfs
196        .name(name() + ".invAddrSwpfs")
197        .desc("Number of software prefetches ignored due to an invalid address");
198
199    lsqBlockedLoads
200        .name(name() + ".blockedLoads")
201        .desc("Number of blocked loads due to partial load-store forwarding");
202
203    lsqRescheduledLoads
204        .name(name() + ".rescheduledLoads")
205        .desc("Number of loads that were rescheduled");
206
207    lsqCacheBlocked
208        .name(name() + ".cacheBlocked")
209        .desc("Number of times an access to memory failed due to the cache being blocked");
210}
211
212template<class Impl>
213void
214LSQUnit<Impl>::clearLQ()
215{
216    loadQueue.clear();
217}
218
219template<class Impl>
220void
221LSQUnit<Impl>::clearSQ()
222{
223    storeQueue.clear();
224}
225
226template<class Impl>
227void
228LSQUnit<Impl>::switchOut()
229{
230    switchedOut = true;
231    for (int i = 0; i < loadQueue.size(); ++i) {
232        assert(!loadQueue[i]);
233        loadQueue[i] = NULL;
234    }
235
236    assert(storesToWB == 0);
237}
238
239template<class Impl>
240void
241LSQUnit<Impl>::takeOverFrom()
242{
243    switchedOut = false;
244    loads = stores = storesToWB = 0;
245
246    loadHead = loadTail = 0;
247
248    storeHead = storeWBIdx = storeTail = 0;
249
250    usedPorts = 0;
251
252    memDepViolator = NULL;
253
254    blockedLoadSeqNum = 0;
255
256    stalled = false;
257    isLoadBlocked = false;
258    loadBlockedHandled = false;
259}
260
261template<class Impl>
262void
263LSQUnit<Impl>::resizeLQ(unsigned size)
264{
265    unsigned size_plus_sentinel = size + 1;
266    assert(size_plus_sentinel >= LQEntries);
267
268    if (size_plus_sentinel > LQEntries) {
269        while (size_plus_sentinel > loadQueue.size()) {
270            DynInstPtr dummy;
271            loadQueue.push_back(dummy);
272            LQEntries++;
273        }
274    } else {
275        LQEntries = size_plus_sentinel;
276    }
277
278}
279
280template<class Impl>
281void
282LSQUnit<Impl>::resizeSQ(unsigned size)
283{
284    unsigned size_plus_sentinel = size + 1;
285    if (size_plus_sentinel > SQEntries) {
286        while (size_plus_sentinel > storeQueue.size()) {
287            SQEntry dummy;
288            storeQueue.push_back(dummy);
289            SQEntries++;
290        }
291    } else {
292        SQEntries = size_plus_sentinel;
293    }
294}
295
296template <class Impl>
297void
298LSQUnit<Impl>::insert(DynInstPtr &inst)
299{
300    assert(inst->isMemRef());
301
302    assert(inst->isLoad() || inst->isStore());
303
304    if (inst->isLoad()) {
305        insertLoad(inst);
306    } else {
307        insertStore(inst);
308    }
309
310    inst->setInLSQ();
311}
312
313template <class Impl>
314void
315LSQUnit<Impl>::insertLoad(DynInstPtr &load_inst)
316{
317    assert((loadTail + 1) % LQEntries != loadHead);
318    assert(loads < LQEntries);
319
320    DPRINTF(LSQUnit, "Inserting load PC %#x, idx:%i [sn:%lli]\n",
321            load_inst->readPC(), loadTail, load_inst->seqNum);
322
323    load_inst->lqIdx = loadTail;
324
325    if (stores == 0) {
326        load_inst->sqIdx = -1;
327    } else {
328        load_inst->sqIdx = storeTail;
329    }
330
331    loadQueue[loadTail] = load_inst;
332
333    incrLdIdx(loadTail);
334
335    ++loads;
336}
337
338template <class Impl>
339void
340LSQUnit<Impl>::insertStore(DynInstPtr &store_inst)
341{
342    // Make sure it is not full before inserting an instruction.
343    assert((storeTail + 1) % SQEntries != storeHead);
344    assert(stores < SQEntries);
345
346    DPRINTF(LSQUnit, "Inserting store PC %#x, idx:%i [sn:%lli]\n",
347            store_inst->readPC(), storeTail, store_inst->seqNum);
348
349    store_inst->sqIdx = storeTail;
350    store_inst->lqIdx = loadTail;
351
352    storeQueue[storeTail] = SQEntry(store_inst);
353
354    incrStIdx(storeTail);
355
356    ++stores;
357}
358
359template <class Impl>
360typename Impl::DynInstPtr
361LSQUnit<Impl>::getMemDepViolator()
362{
363    DynInstPtr temp = memDepViolator;
364
365    memDepViolator = NULL;
366
367    return temp;
368}
369
370template <class Impl>
371unsigned
372LSQUnit<Impl>::numFreeEntries()
373{
374    unsigned free_lq_entries = LQEntries - loads;
375    unsigned free_sq_entries = SQEntries - stores;
376
377    // Both the LQ and SQ entries have an extra dummy entry to differentiate
378    // empty/full conditions.  Subtract 1 from the free entries.
379    if (free_lq_entries < free_sq_entries) {
380        return free_lq_entries - 1;
381    } else {
382        return free_sq_entries - 1;
383    }
384}
385
386template <class Impl>
387int
388LSQUnit<Impl>::numLoadsReady()
389{
390    int load_idx = loadHead;
391    int retval = 0;
392
393    while (load_idx != loadTail) {
394        assert(loadQueue[load_idx]);
395
396        if (loadQueue[load_idx]->readyToIssue()) {
397            ++retval;
398        }
399    }
400
401    return retval;
402}
403
404template <class Impl>
405Fault
406LSQUnit<Impl>::executeLoad(DynInstPtr &inst)
407{
408    using namespace TheISA;
409    // Execute a specific load.
410    Fault load_fault = NoFault;
411
412    DPRINTF(LSQUnit, "Executing load PC %#x, [sn:%lli]\n",
413            inst->readPC(),inst->seqNum);
414
415    assert(!inst->isSquashed());
416
417    load_fault = inst->initiateAcc();
418
419    // If the instruction faulted, then we need to send it along to commit
420    // without the instruction completing.
421    if (load_fault != NoFault) {
422        // Send this instruction to commit, also make sure iew stage
423        // realizes there is activity.
424        // Mark it as executed unless it is an uncached load that
425        // needs to hit the head of commit.
426        if (!(inst->hasRequest() && inst->uncacheable()) ||
427            inst->isAtCommit()) {
428            inst->setExecuted();
429        }
430        iewStage->instToCommit(inst);
431        iewStage->activityThisCycle();
432    } else if (!loadBlocked()) {
433        assert(inst->effAddrValid);
434        int load_idx = inst->lqIdx;
435        incrLdIdx(load_idx);
436        while (load_idx != loadTail) {
437            // Really only need to check loads that have actually executed
438
439            // @todo: For now this is extra conservative, detecting a
440            // violation if the addresses match assuming all accesses
441            // are quad word accesses.
442
443            // @todo: Fix this, magic number being used here
444            if (loadQueue[load_idx]->effAddrValid &&
445                (loadQueue[load_idx]->effAddr >> 8) ==
446                (inst->effAddr >> 8)) {
447                // A load incorrectly passed this load.  Squash and refetch.
448                // For now return a fault to show that it was unsuccessful.
449                DynInstPtr violator = loadQueue[load_idx];
450                if (!memDepViolator ||
451                    (violator->seqNum < memDepViolator->seqNum)) {
452                    memDepViolator = violator;
453                } else {
454                    break;
455                }
456
457                ++lsqMemOrderViolation;
458
459                return genMachineCheckFault();
460            }
461
462            incrLdIdx(load_idx);
463        }
464    }
465
466    return load_fault;
467}
468
469template <class Impl>
470Fault
471LSQUnit<Impl>::executeStore(DynInstPtr &store_inst)
472{
473    using namespace TheISA;
474    // Make sure that a store exists.
475    assert(stores != 0);
476
477    int store_idx = store_inst->sqIdx;
478
479    DPRINTF(LSQUnit, "Executing store PC %#x [sn:%lli]\n",
480            store_inst->readPC(), store_inst->seqNum);
481
482    assert(!store_inst->isSquashed());
483
484    // Check the recently completed loads to see if any match this store's
485    // address.  If so, then we have a memory ordering violation.
486    int load_idx = store_inst->lqIdx;
487
488    Fault store_fault = store_inst->initiateAcc();
489
490    if (storeQueue[store_idx].size == 0) {
491        DPRINTF(LSQUnit,"Fault on Store PC %#x, [sn:%lli],Size = 0\n",
492                store_inst->readPC(),store_inst->seqNum);
493
494        return store_fault;
495    }
496
497    assert(store_fault == NoFault);
498
499    if (store_inst->isStoreConditional()) {
500        // Store conditionals need to set themselves as able to
501        // writeback if we haven't had a fault by here.
502        storeQueue[store_idx].canWB = true;
503
504        ++storesToWB;
505    }
506
507    assert(store_inst->effAddrValid);
508    while (load_idx != loadTail) {
509        // Really only need to check loads that have actually executed
510        // It's safe to check all loads because effAddr is set to
511        // InvalAddr when the dyn inst is created.
512
513        // @todo: For now this is extra conservative, detecting a
514        // violation if the addresses match assuming all accesses
515        // are quad word accesses.
516
517        // @todo: Fix this, magic number being used here
518        if (loadQueue[load_idx]->effAddrValid &&
519            (loadQueue[load_idx]->effAddr >> 8) ==
520            (store_inst->effAddr >> 8)) {
521            // A load incorrectly passed this store.  Squash and refetch.
522            // For now return a fault to show that it was unsuccessful.
523            DynInstPtr violator = loadQueue[load_idx];
524            if (!memDepViolator ||
525                (violator->seqNum < memDepViolator->seqNum)) {
526                memDepViolator = violator;
527            } else {
528                break;
529            }
530
531            ++lsqMemOrderViolation;
532
533            return genMachineCheckFault();
534        }
535
536        incrLdIdx(load_idx);
537    }
538
539    return store_fault;
540}
541
542template <class Impl>
543void
544LSQUnit<Impl>::commitLoad()
545{
546    assert(loadQueue[loadHead]);
547
548    DPRINTF(LSQUnit, "Committing head load instruction, PC %#x\n",
549            loadQueue[loadHead]->readPC());
550
551    loadQueue[loadHead] = NULL;
552
553    incrLdIdx(loadHead);
554
555    --loads;
556}
557
558template <class Impl>
559void
560LSQUnit<Impl>::commitLoads(InstSeqNum &youngest_inst)
561{
562    assert(loads == 0 || loadQueue[loadHead]);
563
564    while (loads != 0 && loadQueue[loadHead]->seqNum <= youngest_inst) {
565        commitLoad();
566    }
567}
568
569template <class Impl>
570void
571LSQUnit<Impl>::commitStores(InstSeqNum &youngest_inst)
572{
573    assert(stores == 0 || storeQueue[storeHead].inst);
574
575    int store_idx = storeHead;
576
577    while (store_idx != storeTail) {
578        assert(storeQueue[store_idx].inst);
579        // Mark any stores that are now committed and have not yet
580        // been marked as able to write back.
581        if (!storeQueue[store_idx].canWB) {
582            if (storeQueue[store_idx].inst->seqNum > youngest_inst) {
583                break;
584            }
585            DPRINTF(LSQUnit, "Marking store as able to write back, PC "
586                    "%#x [sn:%lli]\n",
587                    storeQueue[store_idx].inst->readPC(),
588                    storeQueue[store_idx].inst->seqNum);
589
590            storeQueue[store_idx].canWB = true;
591
592            ++storesToWB;
593        }
594
595        incrStIdx(store_idx);
596    }
597}
598
599template <class Impl>
600void
601LSQUnit<Impl>::writebackStores()
602{
603    while (storesToWB > 0 &&
604           storeWBIdx != storeTail &&
605           storeQueue[storeWBIdx].inst &&
606           storeQueue[storeWBIdx].canWB &&
607           usedPorts < cachePorts) {
608
609        if (isStoreBlocked || lsq->cacheBlocked()) {
610            DPRINTF(LSQUnit, "Unable to write back any more stores, cache"
611                    " is blocked!\n");
612            break;
613        }
614
615        // Store didn't write any data so no need to write it back to
616        // memory.
617        if (storeQueue[storeWBIdx].size == 0) {
618            completeStore(storeWBIdx);
619
620            incrStIdx(storeWBIdx);
621
622            continue;
623        }
624
625        ++usedPorts;
626
627        if (storeQueue[storeWBIdx].inst->isDataPrefetch()) {
628            incrStIdx(storeWBIdx);
629
630            continue;
631        }
632
633        assert(storeQueue[storeWBIdx].req);
634        assert(!storeQueue[storeWBIdx].committed);
635
636        DynInstPtr inst = storeQueue[storeWBIdx].inst;
637
638        Request *req = storeQueue[storeWBIdx].req;
639        storeQueue[storeWBIdx].committed = true;
640
641        assert(!inst->memData);
642        inst->memData = new uint8_t[64];
643
644        memcpy(inst->memData, storeQueue[storeWBIdx].data, req->getSize());
645
646        PacketPtr data_pkt = new Packet(req, MemCmd::WriteReq,
647                                        Packet::Broadcast);
648        data_pkt->dataStatic(inst->memData);
649
650        LSQSenderState *state = new LSQSenderState;
651        state->isLoad = false;
652        state->idx = storeWBIdx;
653        state->inst = inst;
654        data_pkt->senderState = state;
655
656        DPRINTF(LSQUnit, "D-Cache: Writing back store idx:%i PC:%#x "
657                "to Addr:%#x, data:%#x [sn:%lli]\n",
658                storeWBIdx, inst->readPC(),
659                req->getPaddr(), (int)*(inst->memData),
660                inst->seqNum);
661
662        // @todo: Remove this SC hack once the memory system handles it.
663        if (req->isLocked()) {
664            // Disable recording the result temporarily.  Writing to
665            // misc regs normally updates the result, but this is not
666            // the desired behavior when handling store conditionals.
667            inst->recordResult = false;
668            bool success = TheISA::handleLockedWrite(inst.get(), req);
669            inst->recordResult = true;
670
671            if (!success) {
672                // Instantly complete this store.
673                DPRINTF(LSQUnit, "Store conditional [sn:%lli] failed.  "
674                        "Instantly completing it.\n",
675                        inst->seqNum);
676                WritebackEvent *wb = new WritebackEvent(inst, data_pkt, this);
677                wb->schedule(curTick + 1);
678                delete state;
679                completeStore(storeWBIdx);
680                incrStIdx(storeWBIdx);
681                continue;
682            }
683        } else {
684            // Non-store conditionals do not need a writeback.
685            state->noWB = true;
686        }
687
688        if (!dcachePort->sendTiming(data_pkt)) {
689            if (data_pkt->result == Packet::BadAddress) {
690                panic("LSQ sent out a bad address for a completed store!");
691            }
692            // Need to handle becoming blocked on a store.
693            DPRINTF(IEW, "D-Cache became blocked when writing [sn:%lli], will"
694                    "retry later\n",
695                    inst->seqNum);
696            isStoreBlocked = true;
697            ++lsqCacheBlocked;
698            assert(retryPkt == NULL);
699            retryPkt = data_pkt;
700            lsq->setRetryTid(lsqID);
701        } else {
702            storePostSend(data_pkt);
703        }
704    }
705
706    // Not sure this should set it to 0.
707    usedPorts = 0;
708
709    assert(stores >= 0 && storesToWB >= 0);
710}
711
712/*template <class Impl>
713void
714LSQUnit<Impl>::removeMSHR(InstSeqNum seqNum)
715{
716    list<InstSeqNum>::iterator mshr_it = find(mshrSeqNums.begin(),
717                                              mshrSeqNums.end(),
718                                              seqNum);
719
720    if (mshr_it != mshrSeqNums.end()) {
721        mshrSeqNums.erase(mshr_it);
722        DPRINTF(LSQUnit, "Removing MSHR. count = %i\n",mshrSeqNums.size());
723    }
724}*/
725
726template <class Impl>
727void
728LSQUnit<Impl>::squash(const InstSeqNum &squashed_num)
729{
730    DPRINTF(LSQUnit, "Squashing until [sn:%lli]!"
731            "(Loads:%i Stores:%i)\n", squashed_num, loads, stores);
732
733    int load_idx = loadTail;
734    decrLdIdx(load_idx);
735
736    while (loads != 0 && loadQueue[load_idx]->seqNum > squashed_num) {
737        DPRINTF(LSQUnit,"Load Instruction PC %#x squashed, "
738                "[sn:%lli]\n",
739                loadQueue[load_idx]->readPC(),
740                loadQueue[load_idx]->seqNum);
741
742        if (isStalled() && load_idx == stallingLoadIdx) {
743            stalled = false;
744            stallingStoreIsn = 0;
745            stallingLoadIdx = 0;
746        }
747
748        // Clear the smart pointer to make sure it is decremented.
749        loadQueue[load_idx]->setSquashed();
750        loadQueue[load_idx] = NULL;
751        --loads;
752
753        // Inefficient!
754        loadTail = load_idx;
755
756        decrLdIdx(load_idx);
757        ++lsqSquashedLoads;
758    }
759
760    if (isLoadBlocked) {
761        if (squashed_num < blockedLoadSeqNum) {
762            isLoadBlocked = false;
763            loadBlockedHandled = false;
764            blockedLoadSeqNum = 0;
765        }
766    }
767
768    if (memDepViolator && squashed_num < memDepViolator->seqNum) {
769        memDepViolator = NULL;
770    }
771
772    int store_idx = storeTail;
773    decrStIdx(store_idx);
774
775    while (stores != 0 &&
776           storeQueue[store_idx].inst->seqNum > squashed_num) {
777        // Instructions marked as can WB are already committed.
778        if (storeQueue[store_idx].canWB) {
779            break;
780        }
781
782        DPRINTF(LSQUnit,"Store Instruction PC %#x squashed, "
783                "idx:%i [sn:%lli]\n",
784                storeQueue[store_idx].inst->readPC(),
785                store_idx, storeQueue[store_idx].inst->seqNum);
786
787        // I don't think this can happen.  It should have been cleared
788        // by the stalling load.
789        if (isStalled() &&
790            storeQueue[store_idx].inst->seqNum == stallingStoreIsn) {
791            panic("Is stalled should have been cleared by stalling load!\n");
792            stalled = false;
793            stallingStoreIsn = 0;
794        }
795
796        // Clear the smart pointer to make sure it is decremented.
797        storeQueue[store_idx].inst->setSquashed();
798        storeQueue[store_idx].inst = NULL;
799        storeQueue[store_idx].canWB = 0;
800
801        // Must delete request now that it wasn't handed off to
802        // memory.  This is quite ugly.  @todo: Figure out the proper
803        // place to really handle request deletes.
804        delete storeQueue[store_idx].req;
805
806        storeQueue[store_idx].req = NULL;
807        --stores;
808
809        // Inefficient!
810        storeTail = store_idx;
811
812        decrStIdx(store_idx);
813        ++lsqSquashedStores;
814    }
815}
816
817template <class Impl>
818void
819LSQUnit<Impl>::storePostSend(PacketPtr pkt)
820{
821    if (isStalled() &&
822        storeQueue[storeWBIdx].inst->seqNum == stallingStoreIsn) {
823        DPRINTF(LSQUnit, "Unstalling, stalling store [sn:%lli] "
824                "load idx:%i\n",
825                stallingStoreIsn, stallingLoadIdx);
826        stalled = false;
827        stallingStoreIsn = 0;
828        iewStage->replayMemInst(loadQueue[stallingLoadIdx]);
829    }
830
831    if (!storeQueue[storeWBIdx].inst->isStoreConditional()) {
832        // The store is basically completed at this time. This
833        // only works so long as the checker doesn't try to
834        // verify the value in memory for stores.
835        storeQueue[storeWBIdx].inst->setCompleted();
836#if USE_CHECKER
837        if (cpu->checker) {
838            cpu->checker->verify(storeQueue[storeWBIdx].inst);
839        }
840#endif
841    }
842
843    if (pkt->result != Packet::Success) {
844        DPRINTF(LSQUnit,"D-Cache Write Miss on idx:%i!\n",
845                storeWBIdx);
846
847        DPRINTF(Activity, "Active st accessing mem miss [sn:%lli]\n",
848                storeQueue[storeWBIdx].inst->seqNum);
849
850        //mshrSeqNums.push_back(storeQueue[storeWBIdx].inst->seqNum);
851
852        //DPRINTF(LSQUnit, "Added MSHR. count = %i\n",mshrSeqNums.size());
853
854        // @todo: Increment stat here.
855    } else {
856        DPRINTF(LSQUnit,"D-Cache: Write Hit on idx:%i !\n",
857                storeWBIdx);
858
859        DPRINTF(Activity, "Active st accessing mem hit [sn:%lli]\n",
860                storeQueue[storeWBIdx].inst->seqNum);
861    }
862
863    incrStIdx(storeWBIdx);
864}
865
866template <class Impl>
867void
868LSQUnit<Impl>::writeback(DynInstPtr &inst, PacketPtr pkt)
869{
870    iewStage->wakeCPU();
871
872    // Squashed instructions do not need to complete their access.
873    if (inst->isSquashed()) {
874        iewStage->decrWb(inst->seqNum);
875        assert(!inst->isStore());
876        ++lsqIgnoredResponses;
877        return;
878    }
879
880    if (!inst->isExecuted()) {
881        inst->setExecuted();
882
883        // Complete access to copy data to proper place.
884        inst->completeAcc(pkt);
885    }
886
887    // Need to insert instruction into queue to commit
888    iewStage->instToCommit(inst);
889
890    iewStage->activityThisCycle();
891}
892
893template <class Impl>
894void
895LSQUnit<Impl>::completeStore(int store_idx)
896{
897    assert(storeQueue[store_idx].inst);
898    storeQueue[store_idx].completed = true;
899    --storesToWB;
900    // A bit conservative because a store completion may not free up entries,
901    // but hopefully avoids two store completions in one cycle from making
902    // the CPU tick twice.
903    cpu->wakeCPU();
904    cpu->activityThisCycle();
905
906    if (store_idx == storeHead) {
907        do {
908            incrStIdx(storeHead);
909
910            --stores;
911        } while (storeQueue[storeHead].completed &&
912                 storeHead != storeTail);
913
914        iewStage->updateLSQNextCycle = true;
915    }
916
917    DPRINTF(LSQUnit, "Completing store [sn:%lli], idx:%i, store head "
918            "idx:%i\n",
919            storeQueue[store_idx].inst->seqNum, store_idx, storeHead);
920
921    if (isStalled() &&
922        storeQueue[store_idx].inst->seqNum == stallingStoreIsn) {
923        DPRINTF(LSQUnit, "Unstalling, stalling store [sn:%lli] "
924                "load idx:%i\n",
925                stallingStoreIsn, stallingLoadIdx);
926        stalled = false;
927        stallingStoreIsn = 0;
928        iewStage->replayMemInst(loadQueue[stallingLoadIdx]);
929    }
930
931    storeQueue[store_idx].inst->setCompleted();
932
933    // Tell the checker we've completed this instruction.  Some stores
934    // may get reported twice to the checker, but the checker can
935    // handle that case.
936#if USE_CHECKER
937    if (cpu->checker) {
938        cpu->checker->verify(storeQueue[store_idx].inst);
939    }
940#endif
941}
942
943template <class Impl>
944void
945LSQUnit<Impl>::recvRetry()
946{
947    if (isStoreBlocked) {
948        assert(retryPkt != NULL);
949
950        if (dcachePort->sendTiming(retryPkt)) {
951            if (retryPkt->result == Packet::BadAddress) {
952                panic("LSQ sent out a bad address for a completed store!");
953            }
954            storePostSend(retryPkt);
955            retryPkt = NULL;
956            isStoreBlocked = false;
957            lsq->setRetryTid(-1);
958        } else {
959            // Still blocked!
960            ++lsqCacheBlocked;
961            lsq->setRetryTid(lsqID);
962        }
963    } else if (isLoadBlocked) {
964        DPRINTF(LSQUnit, "Loads squash themselves and all younger insts, "
965                "no need to resend packet.\n");
966    } else {
967        DPRINTF(LSQUnit, "Retry received but LSQ is no longer blocked.\n");
968    }
969}
970
971template <class Impl>
972inline void
973LSQUnit<Impl>::incrStIdx(int &store_idx)
974{
975    if (++store_idx >= SQEntries)
976        store_idx = 0;
977}
978
979template <class Impl>
980inline void
981LSQUnit<Impl>::decrStIdx(int &store_idx)
982{
983    if (--store_idx < 0)
984        store_idx += SQEntries;
985}
986
987template <class Impl>
988inline void
989LSQUnit<Impl>::incrLdIdx(int &load_idx)
990{
991    if (++load_idx >= LQEntries)
992        load_idx = 0;
993}
994
995template <class Impl>
996inline void
997LSQUnit<Impl>::decrLdIdx(int &load_idx)
998{
999    if (--load_idx < 0)
1000        load_idx += LQEntries;
1001}
1002
1003template <class Impl>
1004void
1005LSQUnit<Impl>::dumpInsts()
1006{
1007    cprintf("Load store queue: Dumping instructions.\n");
1008    cprintf("Load queue size: %i\n", loads);
1009    cprintf("Load queue: ");
1010
1011    int load_idx = loadHead;
1012
1013    while (load_idx != loadTail && loadQueue[load_idx]) {
1014        cprintf("%#x ", loadQueue[load_idx]->readPC());
1015
1016        incrLdIdx(load_idx);
1017    }
1018
1019    cprintf("Store queue size: %i\n", stores);
1020    cprintf("Store queue: ");
1021
1022    int store_idx = storeHead;
1023
1024    while (store_idx != storeTail && storeQueue[store_idx].inst) {
1025        cprintf("%#x ", storeQueue[store_idx].inst->readPC());
1026
1027        incrStIdx(store_idx);
1028    }
1029
1030    cprintf("\n");
1031}
1032