lsq_unit.hh revision 5999 
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#ifndef __CPU_O3_LSQ_UNIT_HH__
33#define __CPU_O3_LSQ_UNIT_HH__
34
35#include <algorithm>
36#include <cstring>
37#include <map>
38#include <queue>
39
40#include "arch/faults.hh"
41#include "arch/locked_mem.hh"
42#include "config/full_system.hh"
43#include "base/fast_alloc.hh"
44#include "base/hashmap.hh"
45#include "cpu/inst_seq.hh"
46#include "mem/packet.hh"
47#include "mem/port.hh"
48
49class DerivO3CPUParams;
50
51/**
52 * Class that implements the actual LQ and SQ for each specific
53 * thread.  Both are circular queues; load entries are freed upon
54 * committing, while store entries are freed once they writeback. The
55 * LSQUnit tracks if there are memory ordering violations, and also
56 * detects partial load to store forwarding cases (a store only has
57 * part of a load's data) that requires the load to wait until the
58 * store writes back. In the former case it holds onto the instruction
59 * until the dependence unit looks at it, and in the latter it stalls
60 * the LSQ until the store writes back. At that point the load is
61 * replayed.
62 */
63template <class Impl>
64class LSQUnit {
65  protected:
66    typedef TheISA::IntReg IntReg;
67  public:
68    typedef typename Impl::O3CPU O3CPU;
69    typedef typename Impl::DynInstPtr DynInstPtr;
70    typedef typename Impl::CPUPol::IEW IEW;
71    typedef typename Impl::CPUPol::LSQ LSQ;
72    typedef typename Impl::CPUPol::IssueStruct IssueStruct;
73
74  public:
75    /** Constructs an LSQ unit. init() must be called prior to use. */
76    LSQUnit();
77
78    /** Initializes the LSQ unit with the specified number of entries. */
79    void init(O3CPU *cpu_ptr, IEW *iew_ptr, DerivO3CPUParams *params,
80            LSQ *lsq_ptr, unsigned maxLQEntries, unsigned maxSQEntries,
81            unsigned id);
82
83    /** Returns the name of the LSQ unit. */
84    std::string name() const;
85
86    /** Registers statistics. */
87    void regStats();
88
89    /** Sets the pointer to the dcache port. */
90    void setDcachePort(Port *dcache_port);
91
92    /** Switches out LSQ unit. */
93    void switchOut();
94
95    /** Takes over from another CPU's thread. */
96    void takeOverFrom();
97
98    /** Returns if the LSQ is switched out. */
99    bool isSwitchedOut() { return switchedOut; }
100
101    /** Ticks the LSQ unit, which in this case only resets the number of
102     * used cache ports.
103     * @todo: Move the number of used ports up to the LSQ level so it can
104     * be shared by all LSQ units.
105     */
106    void tick() { usedPorts = 0; }
107
108    /** Inserts an instruction. */
109    void insert(DynInstPtr &inst);
110    /** Inserts a load instruction. */
111    void insertLoad(DynInstPtr &load_inst);
112    /** Inserts a store instruction. */
113    void insertStore(DynInstPtr &store_inst);
114
115    /** Executes a load instruction. */
116    Fault executeLoad(DynInstPtr &inst);
117
118    Fault executeLoad(int lq_idx) { panic("Not implemented"); return NoFault; }
119    /** Executes a store instruction. */
120    Fault executeStore(DynInstPtr &inst);
121
122    /** Commits the head load. */
123    void commitLoad();
124    /** Commits loads older than a specific sequence number. */
125    void commitLoads(InstSeqNum &youngest_inst);
126
127    /** Commits stores older than a specific sequence number. */
128    void commitStores(InstSeqNum &youngest_inst);
129
130    /** Writes back stores. */
131    void writebackStores();
132
133    /** Completes the data access that has been returned from the
134     * memory system. */
135    void completeDataAccess(PacketPtr pkt);
136
137    /** Clears all the entries in the LQ. */
138    void clearLQ();
139
140    /** Clears all the entries in the SQ. */
141    void clearSQ();
142
143    /** Resizes the LQ to a given size. */
144    void resizeLQ(unsigned size);
145
146    /** Resizes the SQ to a given size. */
147    void resizeSQ(unsigned size);
148
149    /** Squashes all instructions younger than a specific sequence number. */
150    void squash(const InstSeqNum &squashed_num);
151
152    /** Returns if there is a memory ordering violation. Value is reset upon
153     * call to getMemDepViolator().
154     */
155    bool violation() { return memDepViolator; }
156
157    /** Returns the memory ordering violator. */
158    DynInstPtr getMemDepViolator();
159
160    /** Returns if a load became blocked due to the memory system. */
161    bool loadBlocked()
162    { return isLoadBlocked; }
163
164    /** Clears the signal that a load became blocked. */
165    void clearLoadBlocked()
166    { isLoadBlocked = false; }
167
168    /** Returns if the blocked load was handled. */
169    bool isLoadBlockedHandled()
170    { return loadBlockedHandled; }
171
172    /** Records the blocked load as being handled. */
173    void setLoadBlockedHandled()
174    { loadBlockedHandled = true; }
175
176    /** Returns the number of free entries (min of free LQ and SQ entries). */
177    unsigned numFreeEntries();
178
179    /** Returns the number of loads ready to execute. */
180    int numLoadsReady();
181
182    /** Returns the number of loads in the LQ. */
183    int numLoads() { return loads; }
184
185    /** Returns the number of stores in the SQ. */
186    int numStores() { return stores; }
187
188    /** Returns if either the LQ or SQ is full. */
189    bool isFull() { return lqFull() || sqFull(); }
190
191    /** Returns if the LQ is full. */
192    bool lqFull() { return loads >= (LQEntries - 1); }
193
194    /** Returns if the SQ is full. */
195    bool sqFull() { return stores >= (SQEntries - 1); }
196
197    /** Returns the number of instructions in the LSQ. */
198    unsigned getCount() { return loads + stores; }
199
200    /** Returns if there are any stores to writeback. */
201    bool hasStoresToWB() { return storesToWB; }
202
203    /** Returns the number of stores to writeback. */
204    int numStoresToWB() { return storesToWB; }
205
206    /** Returns if the LSQ unit will writeback on this cycle. */
207    bool willWB() { return storeQueue[storeWBIdx].canWB &&
208                        !storeQueue[storeWBIdx].completed &&
209                        !isStoreBlocked; }
210
211    /** Handles doing the retry. */
212    void recvRetry();
213
214  private:
215    /** Writes back the instruction, sending it to IEW. */
216    void writeback(DynInstPtr &inst, PacketPtr pkt);
217
218    /** Handles completing the send of a store to memory. */
219    void storePostSend(PacketPtr pkt);
220
221    /** Completes the store at the specified index. */
222    void completeStore(int store_idx);
223
224    /** Increments the given store index (circular queue). */
225    inline void incrStIdx(int &store_idx);
226    /** Decrements the given store index (circular queue). */
227    inline void decrStIdx(int &store_idx);
228    /** Increments the given load index (circular queue). */
229    inline void incrLdIdx(int &load_idx);
230    /** Decrements the given load index (circular queue). */
231    inline void decrLdIdx(int &load_idx);
232
233  public:
234    /** Debugging function to dump instructions in the LSQ. */
235    void dumpInsts();
236
237  private:
238    /** Pointer to the CPU. */
239    O3CPU *cpu;
240
241    /** Pointer to the IEW stage. */
242    IEW *iewStage;
243
244    /** Pointer to the LSQ. */
245    LSQ *lsq;
246
247    /** Pointer to the dcache port.  Used only for sending. */
248    Port *dcachePort;
249
250    /** Derived class to hold any sender state the LSQ needs. */
251    class LSQSenderState : public Packet::SenderState, public FastAlloc
252    {
253      public:
254        /** Default constructor. */
255        LSQSenderState()
256            : noWB(false)
257        { }
258
259        /** Instruction who initiated the access to memory. */
260        DynInstPtr inst;
261        /** Whether or not it is a load. */
262        bool isLoad;
263        /** The LQ/SQ index of the instruction. */
264        int idx;
265        /** Whether or not the instruction will need to writeback. */
266        bool noWB;
267    };
268
269    /** Writeback event, specifically for when stores forward data to loads. */
270    class WritebackEvent : public Event {
271      public:
272        /** Constructs a writeback event. */
273        WritebackEvent(DynInstPtr &_inst, PacketPtr pkt, LSQUnit *lsq_ptr);
274
275        /** Processes the writeback event. */
276        void process();
277
278        /** Returns the description of this event. */
279        const char *description() const;
280
281      private:
282        /** Instruction whose results are being written back. */
283        DynInstPtr inst;
284
285        /** The packet that would have been sent to memory. */
286        PacketPtr pkt;
287
288        /** The pointer to the LSQ unit that issued the store. */
289        LSQUnit<Impl> *lsqPtr;
290    };
291
292  public:
293    struct SQEntry {
294        /** Constructs an empty store queue entry. */
295        SQEntry()
296            : inst(NULL), req(NULL), size(0),
297              canWB(0), committed(0), completed(0)
298        {
299            std::memset(data, 0, sizeof(data));
300        }
301
302        /** Constructs a store queue entry for a given instruction. */
303        SQEntry(DynInstPtr &_inst)
304            : inst(_inst), req(NULL), size(0),
305              canWB(0), committed(0), completed(0)
306        {
307            std::memset(data, 0, sizeof(data));
308        }
309
310        /** The store instruction. */
311        DynInstPtr inst;
312        /** The request for the store. */
313        RequestPtr req;
314        /** The size of the store. */
315        int size;
316        /** The store data. */
317        char data[sizeof(IntReg)];
318        /** Whether or not the store can writeback. */
319        bool canWB;
320        /** Whether or not the store is committed. */
321        bool committed;
322        /** Whether or not the store is completed. */
323        bool completed;
324    };
325
326  private:
327    /** The LSQUnit thread id. */
328    unsigned lsqID;
329
330    /** The store queue. */
331    std::vector<SQEntry> storeQueue;
332
333    /** The load queue. */
334    std::vector<DynInstPtr> loadQueue;
335
336    /** The number of LQ entries, plus a sentinel entry (circular queue).
337     *  @todo: Consider having var that records the true number of LQ entries.
338     */
339    unsigned LQEntries;
340    /** The number of SQ entries, plus a sentinel entry (circular queue).
341     *  @todo: Consider having var that records the true number of SQ entries.
342     */
343    unsigned SQEntries;
344
345    /** The number of load instructions in the LQ. */
346    int loads;
347    /** The number of store instructions in the SQ. */
348    int stores;
349    /** The number of store instructions in the SQ waiting to writeback. */
350    int storesToWB;
351
352    /** The index of the head instruction in the LQ. */
353    int loadHead;
354    /** The index of the tail instruction in the LQ. */
355    int loadTail;
356
357    /** The index of the head instruction in the SQ. */
358    int storeHead;
359    /** The index of the first instruction that may be ready to be
360     * written back, and has not yet been written back.
361     */
362    int storeWBIdx;
363    /** The index of the tail instruction in the SQ. */
364    int storeTail;
365
366    /// @todo Consider moving to a more advanced model with write vs read ports
367    /** The number of cache ports available each cycle. */
368    int cachePorts;
369
370    /** The number of used cache ports in this cycle. */
371    int usedPorts;
372
373    /** Is the LSQ switched out. */
374    bool switchedOut;
375
376    //list<InstSeqNum> mshrSeqNums;
377
378    /** Wire to read information from the issue stage time queue. */
379    typename TimeBuffer<IssueStruct>::wire fromIssue;
380
381    /** Whether or not the LSQ is stalled. */
382    bool stalled;
383    /** The store that causes the stall due to partial store to load
384     * forwarding.
385     */
386    InstSeqNum stallingStoreIsn;
387    /** The index of the above store. */
388    int stallingLoadIdx;
389
390    /** The packet that needs to be retried. */
391    PacketPtr retryPkt;
392
393    /** Whehter or not a store is blocked due to the memory system. */
394    bool isStoreBlocked;
395
396    /** Whether or not a load is blocked due to the memory system. */
397    bool isLoadBlocked;
398
399    /** Has the blocked load been handled. */
400    bool loadBlockedHandled;
401
402    /** The sequence number of the blocked load. */
403    InstSeqNum blockedLoadSeqNum;
404
405    /** The oldest load that caused a memory ordering violation. */
406    DynInstPtr memDepViolator;
407
408    // Will also need how many read/write ports the Dcache has.  Or keep track
409    // of that in stage that is one level up, and only call executeLoad/Store
410    // the appropriate number of times.
411    /** Total number of loads forwaded from LSQ stores. */
412    Stats::Scalar lsqForwLoads;
413
414    /** Total number of loads ignored due to invalid addresses. */
415    Stats::Scalar invAddrLoads;
416
417    /** Total number of squashed loads. */
418    Stats::Scalar lsqSquashedLoads;
419
420    /** Total number of responses from the memory system that are
421     * ignored due to the instruction already being squashed. */
422    Stats::Scalar lsqIgnoredResponses;
423
424    /** Tota number of memory ordering violations. */
425    Stats::Scalar lsqMemOrderViolation;
426
427    /** Total number of squashed stores. */
428    Stats::Scalar lsqSquashedStores;
429
430    /** Total number of software prefetches ignored due to invalid addresses. */
431    Stats::Scalar invAddrSwpfs;
432
433    /** Ready loads blocked due to partial store-forwarding. */
434    Stats::Scalar lsqBlockedLoads;
435
436    /** Number of loads that were rescheduled. */
437    Stats::Scalar lsqRescheduledLoads;
438
439    /** Number of times the LSQ is blocked due to the cache. */
440    Stats::Scalar lsqCacheBlocked;
441
442  public:
443    /** Executes the load at the given index. */
444    template <class T>
445    Fault read(Request *req, T &data, int load_idx);
446
447    /** Executes the store at the given index. */
448    template <class T>
449    Fault write(Request *req, T &data, int store_idx);
450
451    /** Returns the index of the head load instruction. */
452    int getLoadHead() { return loadHead; }
453    /** Returns the sequence number of the head load instruction. */
454    InstSeqNum getLoadHeadSeqNum()
455    {
456        if (loadQueue[loadHead]) {
457            return loadQueue[loadHead]->seqNum;
458        } else {
459            return 0;
460        }
461
462    }
463
464    /** Returns the index of the head store instruction. */
465    int getStoreHead() { return storeHead; }
466    /** Returns the sequence number of the head store instruction. */
467    InstSeqNum getStoreHeadSeqNum()
468    {
469        if (storeQueue[storeHead].inst) {
470            return storeQueue[storeHead].inst->seqNum;
471        } else {
472            return 0;
473        }
474
475    }
476
477    /** Returns whether or not the LSQ unit is stalled. */
478    bool isStalled()  { return stalled; }
479};
480
481template <class Impl>
482template <class T>
483Fault
484LSQUnit<Impl>::read(Request *req, T &data, int load_idx)
485{
486    DynInstPtr load_inst = loadQueue[load_idx];
487
488    assert(load_inst);
489
490    assert(!load_inst->isExecuted());
491
492    // Make sure this isn't an uncacheable access
493    // A bit of a hackish way to get uncached accesses to work only if they're
494    // at the head of the LSQ and are ready to commit (at the head of the ROB
495    // too).
496    if (req->isUncacheable() &&
497        (load_idx != loadHead || !load_inst->isAtCommit())) {
498        iewStage->rescheduleMemInst(load_inst);
499        ++lsqRescheduledLoads;
500
501        // Must delete request now that it wasn't handed off to
502        // memory.  This is quite ugly.  @todo: Figure out the proper
503        // place to really handle request deletes.
504        delete req;
505        return TheISA::genMachineCheckFault();
506    }
507
508    // Check the SQ for any previous stores that might lead to forwarding
509    int store_idx = load_inst->sqIdx;
510
511    int store_size = 0;
512
513    DPRINTF(LSQUnit, "Read called, load idx: %i, store idx: %i, "
514            "storeHead: %i addr: %#x\n",
515            load_idx, store_idx, storeHead, req->getPaddr());
516
517    if (req->isLocked()) {
518        // Disable recording the result temporarily.  Writing to misc
519        // regs normally updates the result, but this is not the
520        // desired behavior when handling store conditionals.
521        load_inst->recordResult = false;
522        TheISA::handleLockedRead(load_inst.get(), req);
523        load_inst->recordResult = true;
524    }
525
526    while (store_idx != -1) {
527        // End once we've reached the top of the LSQ
528        if (store_idx == storeWBIdx) {
529            break;
530        }
531
532        // Move the index to one younger
533        if (--store_idx < 0)
534            store_idx += SQEntries;
535
536        assert(storeQueue[store_idx].inst);
537
538        store_size = storeQueue[store_idx].size;
539
540        if (store_size == 0)
541            continue;
542        else if (storeQueue[store_idx].inst->uncacheable())
543            continue;
544
545        assert(storeQueue[store_idx].inst->effAddrValid);
546
547        // Check if the store data is within the lower and upper bounds of
548        // addresses that the request needs.
549        bool store_has_lower_limit =
550            req->getVaddr() >= storeQueue[store_idx].inst->effAddr;
551        bool store_has_upper_limit =
552            (req->getVaddr() + req->getSize()) <=
553            (storeQueue[store_idx].inst->effAddr + store_size);
554        bool lower_load_has_store_part =
555            req->getVaddr() < (storeQueue[store_idx].inst->effAddr +
556                           store_size);
557        bool upper_load_has_store_part =
558            (req->getVaddr() + req->getSize()) >
559            storeQueue[store_idx].inst->effAddr;
560
561        // If the store's data has all of the data needed, we can forward.
562        if ((store_has_lower_limit && store_has_upper_limit)) {
563            // Get shift amount for offset into the store's data.
564            int shift_amt = req->getVaddr() & (store_size - 1);
565
566            memcpy(&data, storeQueue[store_idx].data + shift_amt, sizeof(T));
567
568            assert(!load_inst->memData);
569            load_inst->memData = new uint8_t[64];
570
571            memcpy(load_inst->memData,
572                    storeQueue[store_idx].data + shift_amt, req->getSize());
573
574            DPRINTF(LSQUnit, "Forwarding from store idx %i to load to "
575                    "addr %#x, data %#x\n",
576                    store_idx, req->getVaddr(), data);
577
578            PacketPtr data_pkt = new Packet(req, MemCmd::ReadReq,
579                                            Packet::Broadcast);
580            data_pkt->dataStatic(load_inst->memData);
581
582            WritebackEvent *wb = new WritebackEvent(load_inst, data_pkt, this);
583
584            // We'll say this has a 1 cycle load-store forwarding latency
585            // for now.
586            // @todo: Need to make this a parameter.
587            cpu->schedule(wb, curTick);
588
589            ++lsqForwLoads;
590            return NoFault;
591        } else if ((store_has_lower_limit && lower_load_has_store_part) ||
592                   (store_has_upper_limit && upper_load_has_store_part) ||
593                   (lower_load_has_store_part && upper_load_has_store_part)) {
594            // This is the partial store-load forwarding case where a store
595            // has only part of the load's data.
596
597            // If it's already been written back, then don't worry about
598            // stalling on it.
599            if (storeQueue[store_idx].completed) {
600                panic("Should not check one of these");
601                continue;
602            }
603
604            // Must stall load and force it to retry, so long as it's the oldest
605            // load that needs to do so.
606            if (!stalled ||
607                (stalled &&
608                 load_inst->seqNum <
609                 loadQueue[stallingLoadIdx]->seqNum)) {
610                stalled = true;
611                stallingStoreIsn = storeQueue[store_idx].inst->seqNum;
612                stallingLoadIdx = load_idx;
613            }
614
615            // Tell IQ/mem dep unit that this instruction will need to be
616            // rescheduled eventually
617            iewStage->rescheduleMemInst(load_inst);
618            iewStage->decrWb(load_inst->seqNum);
619            load_inst->clearIssued();
620            ++lsqRescheduledLoads;
621
622            // Do not generate a writeback event as this instruction is not
623            // complete.
624            DPRINTF(LSQUnit, "Load-store forwarding mis-match. "
625                    "Store idx %i to load addr %#x\n",
626                    store_idx, req->getVaddr());
627
628            // Must delete request now that it wasn't handed off to
629            // memory.  This is quite ugly.  @todo: Figure out the
630            // proper place to really handle request deletes.
631            delete req;
632
633            return NoFault;
634        }
635    }
636
637    // If there's no forwarding case, then go access memory
638    DPRINTF(LSQUnit, "Doing memory access for inst [sn:%lli] PC %#x\n",
639            load_inst->seqNum, load_inst->readPC());
640
641    assert(!load_inst->memData);
642    load_inst->memData = new uint8_t[64];
643
644    ++usedPorts;
645
646    // if we the cache is not blocked, do cache access
647    if (!lsq->cacheBlocked()) {
648        PacketPtr data_pkt =
649            new Packet(req,
650                       (req->isLocked() ?
651                        MemCmd::LoadLockedReq : MemCmd::ReadReq),
652                       Packet::Broadcast);
653        data_pkt->dataStatic(load_inst->memData);
654
655        LSQSenderState *state = new LSQSenderState;
656        state->isLoad = true;
657        state->idx = load_idx;
658        state->inst = load_inst;
659        data_pkt->senderState = state;
660
661        if (!dcachePort->sendTiming(data_pkt)) {
662            // Delete state and data packet because a load retry
663            // initiates a pipeline restart; it does not retry.
664            delete state;
665            delete data_pkt->req;
666            delete data_pkt;
667
668            req = NULL;
669
670            // If the access didn't succeed, tell the LSQ by setting
671            // the retry thread id.
672            lsq->setRetryTid(lsqID);
673        }
674    }
675
676    // If the cache was blocked, or has become blocked due to the access,
677    // handle it.
678    if (lsq->cacheBlocked()) {
679        if (req)
680            delete req;
681
682        ++lsqCacheBlocked;
683
684        iewStage->decrWb(load_inst->seqNum);
685        // There's an older load that's already going to squash.
686        if (isLoadBlocked && blockedLoadSeqNum < load_inst->seqNum)
687            return NoFault;
688
689        // Record that the load was blocked due to memory.  This
690        // load will squash all instructions after it, be
691        // refetched, and re-executed.
692        isLoadBlocked = true;
693        loadBlockedHandled = false;
694        blockedLoadSeqNum = load_inst->seqNum;
695        // No fault occurred, even though the interface is blocked.
696        return NoFault;
697    }
698
699    return NoFault;
700}
701
702template <class Impl>
703template <class T>
704Fault
705LSQUnit<Impl>::write(Request *req, T &data, int store_idx)
706{
707    assert(storeQueue[store_idx].inst);
708
709    DPRINTF(LSQUnit, "Doing write to store idx %i, addr %#x data %#x"
710            " | storeHead:%i [sn:%i]\n",
711            store_idx, req->getPaddr(), data, storeHead,
712            storeQueue[store_idx].inst->seqNum);
713
714    storeQueue[store_idx].req = req;
715    storeQueue[store_idx].size = sizeof(T);
716    assert(sizeof(T) <= sizeof(storeQueue[store_idx].data));
717
718    T gData = htog(data);
719    memcpy(storeQueue[store_idx].data, &gData, sizeof(T));
720
721    // This function only writes the data to the store queue, so no fault
722    // can happen here.
723    return NoFault;
724}
725
726#endif // __CPU_O3_LSQ_UNIT_HH__
727