summaryrefslogtreecommitdiff
path: root/src/gpu-compute/gpu_tlb.cc
blob: 05d22dad658169d4832ab46d6225abf04ab9a2a9 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
/*
 * Copyright (c) 2011-2015 Advanced Micro Devices, Inc.
 * All rights reserved.
 *
 * For use for simulation and test purposes only
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are met:
 *
 * 1. Redistributions of source code must retain the above copyright notice,
 * this list of conditions and the following disclaimer.
 *
 * 2. Redistributions in binary form must reproduce the above copyright notice,
 * this list of conditions and the following disclaimer in the documentation
 * and/or other materials provided with the distribution.
 *
 * 3. Neither the name of the copyright holder nor the names of its contributors
 * may be used to endorse or promote products derived from this software
 * without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 *
 * Author: Lisa Hsu
 */

#include "gpu-compute/gpu_tlb.hh"

#include <cmath>
#include <cstring>

#include "arch/x86/faults.hh"
#include "arch/x86/insts/microldstop.hh"
#include "arch/x86/pagetable.hh"
#include "arch/x86/pagetable_walker.hh"
#include "arch/x86/regs/misc.hh"
#include "arch/x86/x86_traits.hh"
#include "base/bitfield.hh"
#include "base/output.hh"
#include "base/trace.hh"
#include "cpu/base.hh"
#include "cpu/thread_context.hh"
#include "debug/GPUPrefetch.hh"
#include "debug/GPUTLB.hh"
#include "mem/packet_access.hh"
#include "mem/page_table.hh"
#include "mem/request.hh"
#include "sim/process.hh"

namespace X86ISA
{

    GpuTLB::GpuTLB(const Params *p)
        : MemObject(p), configAddress(0), size(p->size),
          cleanupEvent([this]{ cleanup(); }, name(), false,
                       Event::Maximum_Pri),
          exitEvent([this]{ exitCallback(); }, name())
    {
        assoc = p->assoc;
        assert(assoc <= size);
        numSets = size/assoc;
        allocationPolicy = p->allocationPolicy;
        hasMemSidePort = false;
        accessDistance = p->accessDistance;
        clock = p->clk_domain->clockPeriod();

        tlb.assign(size, GpuTlbEntry());

        freeList.resize(numSets);
        entryList.resize(numSets);

        for (int set = 0; set < numSets; ++set) {
            for (int way = 0; way < assoc; ++way) {
                int x = set * assoc + way;
                freeList[set].push_back(&tlb.at(x));
            }
        }

        FA = (size == assoc);

        /**
         * @warning: the set-associative version assumes you have a
         * fixed page size of 4KB.
         * If the page size is greather than 4KB (as defined in the
         * TheISA::PageBytes), then there are various issues w/ the current
         * implementation (you'd have the same 8KB page being replicated in
         * different sets etc)
         */
        setMask = numSets - 1;

    #if 0
        // GpuTLB doesn't yet support full system
        walker = p->walker;
        walker->setTLB(this);
    #endif

        maxCoalescedReqs = p->maxOutstandingReqs;

        // Do not allow maxCoalescedReqs to be more than the TLB associativity
        if (maxCoalescedReqs > assoc) {
            maxCoalescedReqs = assoc;
            cprintf("Forcing maxCoalescedReqs to %d (TLB assoc.) \n", assoc);
        }

        outstandingReqs = 0;
        hitLatency = p->hitLatency;
        missLatency1 = p->missLatency1;
        missLatency2 = p->missLatency2;

        // create the slave ports based on the number of connected ports
        for (size_t i = 0; i < p->port_slave_connection_count; ++i) {
            cpuSidePort.push_back(new CpuSidePort(csprintf("%s-port%d",
                                  name(), i), this, i));
        }

        // create the master ports based on the number of connected ports
        for (size_t i = 0; i < p->port_master_connection_count; ++i) {
            memSidePort.push_back(new MemSidePort(csprintf("%s-port%d",
                                  name(), i), this, i));
        }
    }

    // fixme: this is never called?
    GpuTLB::~GpuTLB()
    {
        // make sure all the hash-maps are empty
        assert(translationReturnEvent.empty());
    }

    BaseSlavePort&
    GpuTLB::getSlavePort(const std::string &if_name, PortID idx)
    {
        if (if_name == "slave") {
            if (idx >= static_cast<PortID>(cpuSidePort.size())) {
                panic("TLBCoalescer::getSlavePort: unknown index %d\n", idx);
            }

            return *cpuSidePort[idx];
        } else {
            panic("TLBCoalescer::getSlavePort: unknown port %s\n", if_name);
        }
    }

    BaseMasterPort&
    GpuTLB::getMasterPort(const std::string &if_name, PortID idx)
    {
        if (if_name == "master") {
            if (idx >= static_cast<PortID>(memSidePort.size())) {
                panic("TLBCoalescer::getMasterPort: unknown index %d\n", idx);
            }

            hasMemSidePort = true;

            return *memSidePort[idx];
        } else {
            panic("TLBCoalescer::getMasterPort: unknown port %s\n", if_name);
        }
    }

    GpuTlbEntry*
    GpuTLB::insert(Addr vpn, GpuTlbEntry &entry)
    {
        GpuTlbEntry *newEntry = nullptr;

        /**
         * vpn holds the virtual page address
         * The least significant bits are simply masked
         */
        int set = (vpn >> TheISA::PageShift) & setMask;

        if (!freeList[set].empty()) {
            newEntry = freeList[set].front();
            freeList[set].pop_front();
        } else {
            newEntry = entryList[set].back();
            entryList[set].pop_back();
        }

        *newEntry = entry;
        newEntry->vaddr = vpn;
        entryList[set].push_front(newEntry);

        return newEntry;
    }

    GpuTLB::EntryList::iterator
    GpuTLB::lookupIt(Addr va, bool update_lru)
    {
        int set = (va >> TheISA::PageShift) & setMask;

        if (FA) {
            assert(!set);
        }

        auto entry = entryList[set].begin();
        for (; entry != entryList[set].end(); ++entry) {
            int page_size = (*entry)->size();

            if ((*entry)->vaddr <= va && (*entry)->vaddr + page_size > va) {
                DPRINTF(GPUTLB, "Matched vaddr %#x to entry starting at %#x "
                        "with size %#x.\n", va, (*entry)->vaddr, page_size);

                if (update_lru) {
                    entryList[set].push_front(*entry);
                    entryList[set].erase(entry);
                    entry = entryList[set].begin();
                }

                break;
            }
        }

        return entry;
    }

    GpuTlbEntry*
    GpuTLB::lookup(Addr va, bool update_lru)
    {
        int set = (va >> TheISA::PageShift) & setMask;

        auto entry = lookupIt(va, update_lru);

        if (entry == entryList[set].end())
            return nullptr;
        else
            return *entry;
    }

    void
    GpuTLB::invalidateAll()
    {
        DPRINTF(GPUTLB, "Invalidating all entries.\n");

        for (int i = 0; i < numSets; ++i) {
            while (!entryList[i].empty()) {
                GpuTlbEntry *entry = entryList[i].front();
                entryList[i].pop_front();
                freeList[i].push_back(entry);
            }
        }
    }

    void
    GpuTLB::setConfigAddress(uint32_t addr)
    {
        configAddress = addr;
    }

    void
    GpuTLB::invalidateNonGlobal()
    {
        DPRINTF(GPUTLB, "Invalidating all non global entries.\n");

        for (int i = 0; i < numSets; ++i) {
            for (auto entryIt = entryList[i].begin();
                 entryIt != entryList[i].end();) {
                if (!(*entryIt)->global) {
                    freeList[i].push_back(*entryIt);
                    entryList[i].erase(entryIt++);
                } else {
                    ++entryIt;
                }
            }
        }
    }

    void
    GpuTLB::demapPage(Addr va, uint64_t asn)
    {

        int set = (va >> TheISA::PageShift) & setMask;
        auto entry = lookupIt(va, false);

        if (entry != entryList[set].end()) {
            freeList[set].push_back(*entry);
            entryList[set].erase(entry);
        }
    }

    Fault
    GpuTLB::translateInt(RequestPtr req, ThreadContext *tc)
    {
        DPRINTF(GPUTLB, "Addresses references internal memory.\n");
        Addr vaddr = req->getVaddr();
        Addr prefix = (vaddr >> 3) & IntAddrPrefixMask;

        if (prefix == IntAddrPrefixCPUID) {
            panic("CPUID memory space not yet implemented!\n");
        } else if (prefix == IntAddrPrefixMSR) {
            vaddr = vaddr >> 3;
            req->setFlags(Request::MMAPPED_IPR);
            Addr regNum = 0;

            switch (vaddr & ~IntAddrPrefixMask) {
              case 0x10:
                regNum = MISCREG_TSC;
                break;
              case 0x1B:
                regNum = MISCREG_APIC_BASE;
                break;
              case 0xFE:
                regNum = MISCREG_MTRRCAP;
                break;
              case 0x174:
                regNum = MISCREG_SYSENTER_CS;
                break;
              case 0x175:
                regNum = MISCREG_SYSENTER_ESP;
                break;
              case 0x176:
                regNum = MISCREG_SYSENTER_EIP;
                break;
              case 0x179:
                regNum = MISCREG_MCG_CAP;
                break;
              case 0x17A:
                regNum = MISCREG_MCG_STATUS;
                break;
              case 0x17B:
                regNum = MISCREG_MCG_CTL;
                break;
              case 0x1D9:
                regNum = MISCREG_DEBUG_CTL_MSR;
                break;
              case 0x1DB:
                regNum = MISCREG_LAST_BRANCH_FROM_IP;
                break;
              case 0x1DC:
                regNum = MISCREG_LAST_BRANCH_TO_IP;
                break;
              case 0x1DD:
                regNum = MISCREG_LAST_EXCEPTION_FROM_IP;
                break;
              case 0x1DE:
                regNum = MISCREG_LAST_EXCEPTION_TO_IP;
                break;
              case 0x200:
                regNum = MISCREG_MTRR_PHYS_BASE_0;
                break;
              case 0x201:
                regNum = MISCREG_MTRR_PHYS_MASK_0;
                break;
              case 0x202:
                regNum = MISCREG_MTRR_PHYS_BASE_1;
                break;
              case 0x203:
                regNum = MISCREG_MTRR_PHYS_MASK_1;
                break;
              case 0x204:
                regNum = MISCREG_MTRR_PHYS_BASE_2;
                break;
              case 0x205:
                regNum = MISCREG_MTRR_PHYS_MASK_2;
                break;
              case 0x206:
                regNum = MISCREG_MTRR_PHYS_BASE_3;
                break;
              case 0x207:
                regNum = MISCREG_MTRR_PHYS_MASK_3;
                break;
              case 0x208:
                regNum = MISCREG_MTRR_PHYS_BASE_4;
                break;
              case 0x209:
                regNum = MISCREG_MTRR_PHYS_MASK_4;
                break;
              case 0x20A:
                regNum = MISCREG_MTRR_PHYS_BASE_5;
                break;
              case 0x20B:
                regNum = MISCREG_MTRR_PHYS_MASK_5;
                break;
              case 0x20C:
                regNum = MISCREG_MTRR_PHYS_BASE_6;
                break;
              case 0x20D:
                regNum = MISCREG_MTRR_PHYS_MASK_6;
                break;
              case 0x20E:
                regNum = MISCREG_MTRR_PHYS_BASE_7;
                break;
              case 0x20F:
                regNum = MISCREG_MTRR_PHYS_MASK_7;
                break;
              case 0x250:
                regNum = MISCREG_MTRR_FIX_64K_00000;
                break;
              case 0x258:
                regNum = MISCREG_MTRR_FIX_16K_80000;
                break;
              case 0x259:
                regNum = MISCREG_MTRR_FIX_16K_A0000;
                break;
              case 0x268:
                regNum = MISCREG_MTRR_FIX_4K_C0000;
                break;
              case 0x269:
                regNum = MISCREG_MTRR_FIX_4K_C8000;
                break;
              case 0x26A:
                regNum = MISCREG_MTRR_FIX_4K_D0000;
                break;
              case 0x26B:
                regNum = MISCREG_MTRR_FIX_4K_D8000;
                break;
              case 0x26C:
                regNum = MISCREG_MTRR_FIX_4K_E0000;
                break;
              case 0x26D:
                regNum = MISCREG_MTRR_FIX_4K_E8000;
                break;
              case 0x26E:
                regNum = MISCREG_MTRR_FIX_4K_F0000;
                break;
              case 0x26F:
                regNum = MISCREG_MTRR_FIX_4K_F8000;
                break;
              case 0x277:
                regNum = MISCREG_PAT;
                break;
              case 0x2FF:
                regNum = MISCREG_DEF_TYPE;
                break;
              case 0x400:
                regNum = MISCREG_MC0_CTL;
                break;
              case 0x404:
                regNum = MISCREG_MC1_CTL;
                break;
              case 0x408:
                regNum = MISCREG_MC2_CTL;
                break;
              case 0x40C:
                regNum = MISCREG_MC3_CTL;
                break;
              case 0x410:
                regNum = MISCREG_MC4_CTL;
                break;
              case 0x414:
                regNum = MISCREG_MC5_CTL;
                break;
              case 0x418:
                regNum = MISCREG_MC6_CTL;
                break;
              case 0x41C:
                regNum = MISCREG_MC7_CTL;
                break;
              case 0x401:
                regNum = MISCREG_MC0_STATUS;
                break;
              case 0x405:
                regNum = MISCREG_MC1_STATUS;
                break;
              case 0x409:
                regNum = MISCREG_MC2_STATUS;
                break;
              case 0x40D:
                regNum = MISCREG_MC3_STATUS;
                break;
              case 0x411:
                regNum = MISCREG_MC4_STATUS;
                break;
              case 0x415:
                regNum = MISCREG_MC5_STATUS;
                break;
              case 0x419:
                regNum = MISCREG_MC6_STATUS;
                break;
              case 0x41D:
                regNum = MISCREG_MC7_STATUS;
                break;
              case 0x402:
                regNum = MISCREG_MC0_ADDR;
                break;
              case 0x406:
                regNum = MISCREG_MC1_ADDR;
                break;
              case 0x40A:
                regNum = MISCREG_MC2_ADDR;
                break;
              case 0x40E:
                regNum = MISCREG_MC3_ADDR;
                break;
              case 0x412:
                regNum = MISCREG_MC4_ADDR;
                break;
              case 0x416:
                regNum = MISCREG_MC5_ADDR;
                break;
              case 0x41A:
                regNum = MISCREG_MC6_ADDR;
                break;
              case 0x41E:
                regNum = MISCREG_MC7_ADDR;
                break;
              case 0x403:
                regNum = MISCREG_MC0_MISC;
                break;
              case 0x407:
                regNum = MISCREG_MC1_MISC;
                break;
              case 0x40B:
                regNum = MISCREG_MC2_MISC;
                break;
              case 0x40F:
                regNum = MISCREG_MC3_MISC;
                break;
              case 0x413:
                regNum = MISCREG_MC4_MISC;
                break;
              case 0x417:
                regNum = MISCREG_MC5_MISC;
                break;
              case 0x41B:
                regNum = MISCREG_MC6_MISC;
                break;
              case 0x41F:
                regNum = MISCREG_MC7_MISC;
                break;
              case 0xC0000080:
                regNum = MISCREG_EFER;
                break;
              case 0xC0000081:
                regNum = MISCREG_STAR;
                break;
              case 0xC0000082:
                regNum = MISCREG_LSTAR;
                break;
              case 0xC0000083:
                regNum = MISCREG_CSTAR;
                break;
              case 0xC0000084:
                regNum = MISCREG_SF_MASK;
                break;
              case 0xC0000100:
                regNum = MISCREG_FS_BASE;
                break;
              case 0xC0000101:
                regNum = MISCREG_GS_BASE;
                break;
              case 0xC0000102:
                regNum = MISCREG_KERNEL_GS_BASE;
                break;
              case 0xC0000103:
                regNum = MISCREG_TSC_AUX;
                break;
              case 0xC0010000:
                regNum = MISCREG_PERF_EVT_SEL0;
                break;
              case 0xC0010001:
                regNum = MISCREG_PERF_EVT_SEL1;
                break;
              case 0xC0010002:
                regNum = MISCREG_PERF_EVT_SEL2;
                break;
              case 0xC0010003:
                regNum = MISCREG_PERF_EVT_SEL3;
                break;
              case 0xC0010004:
                regNum = MISCREG_PERF_EVT_CTR0;
                break;
              case 0xC0010005:
                regNum = MISCREG_PERF_EVT_CTR1;
                break;
              case 0xC0010006:
                regNum = MISCREG_PERF_EVT_CTR2;
                break;
              case 0xC0010007:
                regNum = MISCREG_PERF_EVT_CTR3;
                break;
              case 0xC0010010:
                regNum = MISCREG_SYSCFG;
                break;
              case 0xC0010016:
                regNum = MISCREG_IORR_BASE0;
                break;
              case 0xC0010017:
                regNum = MISCREG_IORR_BASE1;
                break;
              case 0xC0010018:
                regNum = MISCREG_IORR_MASK0;
                break;
              case 0xC0010019:
                regNum = MISCREG_IORR_MASK1;
                break;
              case 0xC001001A:
                regNum = MISCREG_TOP_MEM;
                break;
              case 0xC001001D:
                regNum = MISCREG_TOP_MEM2;
                break;
              case 0xC0010114:
                regNum = MISCREG_VM_CR;
                break;
              case 0xC0010115:
                regNum = MISCREG_IGNNE;
                break;
              case 0xC0010116:
                regNum = MISCREG_SMM_CTL;
                break;
              case 0xC0010117:
                regNum = MISCREG_VM_HSAVE_PA;
                break;
              default:
                return std::make_shared<GeneralProtection>(0);
            }
            //The index is multiplied by the size of a MiscReg so that
            //any memory dependence calculations will not see these as
            //overlapping.
            req->setPaddr(regNum * sizeof(MiscReg));
            return NoFault;
        } else if (prefix == IntAddrPrefixIO) {
            // TODO If CPL > IOPL or in virtual mode, check the I/O permission
            // bitmap in the TSS.

            Addr IOPort = vaddr & ~IntAddrPrefixMask;
            // Make sure the address fits in the expected 16 bit IO address
            // space.
            assert(!(IOPort & ~0xFFFF));

            if (IOPort == 0xCF8 && req->getSize() == 4) {
                req->setFlags(Request::MMAPPED_IPR);
                req->setPaddr(MISCREG_PCI_CONFIG_ADDRESS * sizeof(MiscReg));
            } else if ((IOPort & ~mask(2)) == 0xCFC) {
                req->setFlags(Request::UNCACHEABLE);

                Addr configAddress =
                    tc->readMiscRegNoEffect(MISCREG_PCI_CONFIG_ADDRESS);

                if (bits(configAddress, 31, 31)) {
                    req->setPaddr(PhysAddrPrefixPciConfig |
                                  mbits(configAddress, 30, 2) |
                                  (IOPort & mask(2)));
                } else {
                    req->setPaddr(PhysAddrPrefixIO | IOPort);
                }
            } else {
                req->setFlags(Request::UNCACHEABLE);
                req->setPaddr(PhysAddrPrefixIO | IOPort);
            }
            return NoFault;
        } else {
            panic("Access to unrecognized internal address space %#x.\n",
                  prefix);
        }
    }

    /**
     * TLB_lookup will only perform a TLB lookup returning true on a TLB hit
     * and false on a TLB miss.
     * Many of the checks about different modes have been converted to
     * assertions, since these parts of the code are not really used.
     * On a hit it will update the LRU stack.
     */
    bool
    GpuTLB::tlbLookup(RequestPtr req, ThreadContext *tc, bool update_stats)
    {
        bool tlb_hit = false;
    #ifndef NDEBUG
        uint32_t flags = req->getFlags();
        int seg = flags & SegmentFlagMask;
    #endif

        assert(seg != SEGMENT_REG_MS);
        Addr vaddr = req->getVaddr();
        DPRINTF(GPUTLB, "TLB Lookup for vaddr %#x.\n", vaddr);
        HandyM5Reg m5Reg = tc->readMiscRegNoEffect(MISCREG_M5_REG);

        if (m5Reg.prot) {
            DPRINTF(GPUTLB, "In protected mode.\n");
            // make sure we are in 64-bit mode
            assert(m5Reg.mode == LongMode);

            // If paging is enabled, do the translation.
            if (m5Reg.paging) {
                DPRINTF(GPUTLB, "Paging enabled.\n");
                //update LRU stack on a hit
                GpuTlbEntry *entry = lookup(vaddr, true);

                if (entry)
                    tlb_hit = true;

                if (!update_stats) {
                    // functional tlb access for memory initialization
                    // i.e., memory seeding or instr. seeding -> don't update
                    // TLB and stats
                    return tlb_hit;
                }

                localNumTLBAccesses++;

                if (!entry) {
                    localNumTLBMisses++;
                } else {
                    localNumTLBHits++;
                }
            }
        }

        return tlb_hit;
    }

    Fault
    GpuTLB::translate(RequestPtr req, ThreadContext *tc,
                      Translation *translation, Mode mode,
                      bool &delayedResponse, bool timing, int &latency)
    {
        uint32_t flags = req->getFlags();
        int seg = flags & SegmentFlagMask;
        bool storeCheck = flags & (StoreCheck << FlagShift);

        // If this is true, we're dealing with a request
        // to a non-memory address space.
        if (seg == SEGMENT_REG_MS) {
            return translateInt(req, tc);
        }

        delayedResponse = false;
        Addr vaddr = req->getVaddr();
        DPRINTF(GPUTLB, "Translating vaddr %#x.\n", vaddr);

        HandyM5Reg m5Reg = tc->readMiscRegNoEffect(MISCREG_M5_REG);

        // If protected mode has been enabled...
        if (m5Reg.prot) {
            DPRINTF(GPUTLB, "In protected mode.\n");
            // If we're not in 64-bit mode, do protection/limit checks
            if (m5Reg.mode != LongMode) {
                DPRINTF(GPUTLB, "Not in long mode. Checking segment "
                        "protection.\n");

                // Check for a null segment selector.
                if (!(seg == SEGMENT_REG_TSG || seg == SYS_SEGMENT_REG_IDTR ||
                    seg == SEGMENT_REG_HS || seg == SEGMENT_REG_LS)
                    && !tc->readMiscRegNoEffect(MISCREG_SEG_SEL(seg))) {
                    return std::make_shared<GeneralProtection>(0);
                }

                bool expandDown = false;
                SegAttr attr = tc->readMiscRegNoEffect(MISCREG_SEG_ATTR(seg));

                if (seg >= SEGMENT_REG_ES && seg <= SEGMENT_REG_HS) {
                    if (!attr.writable && (mode == BaseTLB::Write ||
                        storeCheck))
                        return std::make_shared<GeneralProtection>(0);

                    if (!attr.readable && mode == BaseTLB::Read)
                        return std::make_shared<GeneralProtection>(0);

                    expandDown = attr.expandDown;

                }

                Addr base = tc->readMiscRegNoEffect(MISCREG_SEG_BASE(seg));
                Addr limit = tc->readMiscRegNoEffect(MISCREG_SEG_LIMIT(seg));
                // This assumes we're not in 64 bit mode. If we were, the
                // default address size is 64 bits, overridable to 32.
                int size = 32;
                bool sizeOverride = (flags & (AddrSizeFlagBit << FlagShift));
                SegAttr csAttr = tc->readMiscRegNoEffect(MISCREG_CS_ATTR);

                if ((csAttr.defaultSize && sizeOverride) ||
                    (!csAttr.defaultSize && !sizeOverride)) {
                    size = 16;
                }

                Addr offset = bits(vaddr - base, size - 1, 0);
                Addr endOffset = offset + req->getSize() - 1;

                if (expandDown) {
                    DPRINTF(GPUTLB, "Checking an expand down segment.\n");
                    warn_once("Expand down segments are untested.\n");

                    if (offset <= limit || endOffset <= limit)
                        return std::make_shared<GeneralProtection>(0);
                } else {
                    if (offset > limit || endOffset > limit)
                        return std::make_shared<GeneralProtection>(0);
                }
            }

            // If paging is enabled, do the translation.
            if (m5Reg.paging) {
                DPRINTF(GPUTLB, "Paging enabled.\n");
                // The vaddr already has the segment base applied.
                GpuTlbEntry *entry = lookup(vaddr);
                localNumTLBAccesses++;

                if (!entry) {
                    localNumTLBMisses++;
                    if (timing) {
                        latency = missLatency1;
                    }

                    if (FullSystem) {
                        fatal("GpuTLB doesn't support full-system mode\n");
                    } else {
                        DPRINTF(GPUTLB, "Handling a TLB miss for address %#x "
                                "at pc %#x.\n", vaddr, tc->instAddr());

                        Process *p = tc->getProcessPtr();
                        const EmulationPageTable::Entry *pte =
                            p->pTable->lookup(vaddr);

                        if (!pte && mode != BaseTLB::Execute) {
                            // penalize a "page fault" more
                            if (timing)
                                latency += missLatency2;

                            if (p->fixupStackFault(vaddr))
                                pte = p->pTable->lookup(vaddr);
                        }

                        if (!pte) {
                            return std::make_shared<PageFault>(vaddr, true,
                                                               mode, true,
                                                               false);
                        } else {
                            Addr alignedVaddr = p->pTable->pageAlign(vaddr);

                            DPRINTF(GPUTLB, "Mapping %#x to %#x\n",
                                    alignedVaddr, pte->paddr);

                            GpuTlbEntry gpuEntry(
                                p->pTable->pid(), alignedVaddr,
                                pte->paddr, true);
                            entry = insert(alignedVaddr, gpuEntry);
                        }

                        DPRINTF(GPUTLB, "Miss was serviced.\n");
                    }
                } else {
                    localNumTLBHits++;

                    if (timing) {
                        latency = hitLatency;
                    }
                }

                // Do paging protection checks.
                bool inUser = (m5Reg.cpl == 3 &&
                               !(flags & (CPL0FlagBit << FlagShift)));

                CR0 cr0 = tc->readMiscRegNoEffect(MISCREG_CR0);
                bool badWrite = (!entry->writable && (inUser || cr0.wp));

                if ((inUser && !entry->user) || (mode == BaseTLB::Write &&
                     badWrite)) {
                    // The page must have been present to get into the TLB in
                    // the first place. We'll assume the reserved bits are
                    // fine even though we're not checking them.
                    return std::make_shared<PageFault>(vaddr, true, mode,
                                                       inUser, false);
                }

                if (storeCheck && badWrite) {
                    // This would fault if this were a write, so return a page
                    // fault that reflects that happening.
                    return std::make_shared<PageFault>(vaddr, true,
                                                       BaseTLB::Write,
                                                       inUser, false);
                }


                DPRINTF(GPUTLB, "Entry found with paddr %#x, doing protection "
                        "checks.\n", entry->paddr);

                int page_size = entry->size();
                Addr paddr = entry->paddr | (vaddr & (page_size - 1));
                DPRINTF(GPUTLB, "Translated %#x -> %#x.\n", vaddr, paddr);
                req->setPaddr(paddr);

                if (entry->uncacheable)
                    req->setFlags(Request::UNCACHEABLE);
            } else {
                //Use the address which already has segmentation applied.
                DPRINTF(GPUTLB, "Paging disabled.\n");
                DPRINTF(GPUTLB, "Translated %#x -> %#x.\n", vaddr, vaddr);
                req->setPaddr(vaddr);
            }
        } else {
            // Real mode
            DPRINTF(GPUTLB, "In real mode.\n");
            DPRINTF(GPUTLB, "Translated %#x -> %#x.\n", vaddr, vaddr);
            req->setPaddr(vaddr);
        }

        // Check for an access to the local APIC
        if (FullSystem) {
            LocalApicBase localApicBase =
                tc->readMiscRegNoEffect(MISCREG_APIC_BASE);

            Addr baseAddr = localApicBase.base * PageBytes;
            Addr paddr = req->getPaddr();

            if (baseAddr <= paddr && baseAddr + PageBytes > paddr) {
                // Force the access to be uncacheable.
                req->setFlags(Request::UNCACHEABLE);
                req->setPaddr(x86LocalAPICAddress(tc->contextId(),
                                                  paddr - baseAddr));
            }
        }

        return NoFault;
    };

    Fault
    GpuTLB::translateAtomic(RequestPtr req, ThreadContext *tc, Mode mode,
                            int &latency)
    {
        bool delayedResponse;

        return GpuTLB::translate(req, tc, nullptr, mode, delayedResponse, false,
                                 latency);
    }

    void
    GpuTLB::translateTiming(RequestPtr req, ThreadContext *tc,
            Translation *translation, Mode mode, int &latency)
    {
        bool delayedResponse;
        assert(translation);

        Fault fault = GpuTLB::translate(req, tc, translation, mode,
                                        delayedResponse, true, latency);

        if (!delayedResponse)
            translation->finish(fault, req, tc, mode);
    }

    Walker*
    GpuTLB::getWalker()
    {
        return walker;
    }


    void
    GpuTLB::serialize(CheckpointOut &cp) const
    {
    }

    void
    GpuTLB::unserialize(CheckpointIn &cp)
    {
    }

    void
    GpuTLB::regStats()
    {
        MemObject::regStats();

        localNumTLBAccesses
            .name(name() + ".local_TLB_accesses")
            .desc("Number of TLB accesses")
            ;

        localNumTLBHits
            .name(name() + ".local_TLB_hits")
            .desc("Number of TLB hits")
            ;

        localNumTLBMisses
            .name(name() + ".local_TLB_misses")
            .desc("Number of TLB misses")
            ;

        localTLBMissRate
            .name(name() + ".local_TLB_miss_rate")
            .desc("TLB miss rate")
            ;

        accessCycles
            .name(name() + ".access_cycles")
            .desc("Cycles spent accessing this TLB level")
            ;

        pageTableCycles
            .name(name() + ".page_table_cycles")
            .desc("Cycles spent accessing the page table")
            ;

        localTLBMissRate = 100 * localNumTLBMisses / localNumTLBAccesses;

        numUniquePages
            .name(name() + ".unique_pages")
            .desc("Number of unique pages touched")
            ;

        localCycles
            .name(name() + ".local_cycles")
            .desc("Number of cycles spent in queue for all incoming reqs")
            ;

        localLatency
            .name(name() + ".local_latency")
            .desc("Avg. latency over incoming coalesced reqs")
            ;

        localLatency = localCycles / localNumTLBAccesses;

        globalNumTLBAccesses
            .name(name() + ".global_TLB_accesses")
            .desc("Number of TLB accesses")
            ;

        globalNumTLBHits
            .name(name() + ".global_TLB_hits")
            .desc("Number of TLB hits")
            ;

        globalNumTLBMisses
            .name(name() + ".global_TLB_misses")
            .desc("Number of TLB misses")
            ;

        globalTLBMissRate
            .name(name() + ".global_TLB_miss_rate")
            .desc("TLB miss rate")
            ;

        globalTLBMissRate = 100 * globalNumTLBMisses / globalNumTLBAccesses;

        avgReuseDistance
            .name(name() + ".avg_reuse_distance")
            .desc("avg. reuse distance over all pages (in ticks)")
            ;

    }

    /**
     * Do the TLB lookup for this coalesced request and schedule
     * another event <TLB access latency> cycles later.
     */

    void
    GpuTLB::issueTLBLookup(PacketPtr pkt)
    {
        assert(pkt);
        assert(pkt->senderState);

        Addr virt_page_addr = roundDown(pkt->req->getVaddr(),
                                        TheISA::PageBytes);

        TranslationState *sender_state =
                safe_cast<TranslationState*>(pkt->senderState);

        bool update_stats = !sender_state->prefetch;
        ThreadContext * tmp_tc = sender_state->tc;

        DPRINTF(GPUTLB, "Translation req. for virt. page addr %#x\n",
                virt_page_addr);

        int req_cnt = sender_state->reqCnt.back();

        if (update_stats) {
            accessCycles -= (curTick() * req_cnt);
            localCycles -= curTick();
            updatePageFootprint(virt_page_addr);
            globalNumTLBAccesses += req_cnt;
        }

        tlbOutcome lookup_outcome = TLB_MISS;
        RequestPtr tmp_req = pkt->req;

        // Access the TLB and figure out if it's a hit or a miss.
        bool success = tlbLookup(tmp_req, tmp_tc, update_stats);

        if (success) {
            lookup_outcome = TLB_HIT;
            // Put the entry in SenderState
            GpuTlbEntry *entry = lookup(tmp_req->getVaddr(), false);
            assert(entry);

            sender_state->tlbEntry =
                new GpuTlbEntry(0, entry->vaddr, entry->paddr, entry->valid);

            if (update_stats) {
                // the reqCnt has an entry per level, so its size tells us
                // which level we are in
                sender_state->hitLevel = sender_state->reqCnt.size();
                globalNumTLBHits += req_cnt;
            }
        } else {
            if (update_stats)
                globalNumTLBMisses += req_cnt;
        }

        /*
         * We now know the TLB lookup outcome (if it's a hit or a miss), as well
         * as the TLB access latency.
         *
         * We create and schedule a new TLBEvent which will help us take the
         * appropriate actions (e.g., update TLB on a hit, send request to lower
         * level TLB on a miss, or start a page walk if this was the last-level
         * TLB)
         */
        TLBEvent *tlb_event =
            new TLBEvent(this, virt_page_addr, lookup_outcome, pkt);

        if (translationReturnEvent.count(virt_page_addr)) {
            panic("Virtual Page Address %#x already has a return event\n",
                  virt_page_addr);
        }

        translationReturnEvent[virt_page_addr] = tlb_event;
        assert(tlb_event);

        DPRINTF(GPUTLB, "schedule translationReturnEvent @ curTick %d\n",
                curTick() + this->ticks(hitLatency));

        schedule(tlb_event, curTick() + this->ticks(hitLatency));
    }

    GpuTLB::TLBEvent::TLBEvent(GpuTLB* _tlb, Addr _addr, tlbOutcome tlb_outcome,
                               PacketPtr _pkt)
        : Event(CPU_Tick_Pri), tlb(_tlb), virtPageAddr(_addr),
        outcome(tlb_outcome), pkt(_pkt)
    {
    }

    /**
     * Do Paging protection checks. If we encounter a page fault, then
     * an assertion is fired.
     */
    void
    GpuTLB::pagingProtectionChecks(ThreadContext *tc, PacketPtr pkt,
            GpuTlbEntry * tlb_entry, Mode mode)
    {
        HandyM5Reg m5Reg = tc->readMiscRegNoEffect(MISCREG_M5_REG);
        uint32_t flags = pkt->req->getFlags();
        bool storeCheck = flags & (StoreCheck << FlagShift);

        // Do paging protection checks.
        bool inUser = (m5Reg.cpl == 3 && !(flags & (CPL0FlagBit << FlagShift)));
        CR0 cr0 = tc->readMiscRegNoEffect(MISCREG_CR0);

        bool badWrite = (!tlb_entry->writable && (inUser || cr0.wp));

        if ((inUser && !tlb_entry->user) ||
            (mode == BaseTLB::Write && badWrite)) {
           // The page must have been present to get into the TLB in
           // the first place. We'll assume the reserved bits are
           // fine even though we're not checking them.
           assert(false);
        }

        if (storeCheck && badWrite) {
           // This would fault if this were a write, so return a page
           // fault that reflects that happening.
           assert(false);
        }
    }

    /**
     * handleTranslationReturn is called on a TLB hit,
     * when a TLB miss returns or when a page fault returns.
     * The latter calls handelHit with TLB miss as tlbOutcome.
     */
    void
    GpuTLB::handleTranslationReturn(Addr virt_page_addr, tlbOutcome tlb_outcome,
            PacketPtr pkt)
    {

        assert(pkt);
        Addr vaddr = pkt->req->getVaddr();

        TranslationState *sender_state =
            safe_cast<TranslationState*>(pkt->senderState);

        ThreadContext *tc = sender_state->tc;
        Mode mode = sender_state->tlbMode;

        GpuTlbEntry *local_entry, *new_entry;

        if (tlb_outcome == TLB_HIT) {
            DPRINTF(GPUTLB, "Translation Done - TLB Hit for addr %#x\n", vaddr);
            local_entry = sender_state->tlbEntry;
        } else {
            DPRINTF(GPUTLB, "Translation Done - TLB Miss for addr %#x\n",
                    vaddr);

            // We are returning either from a page walk or from a hit at a lower
            // TLB level. The senderState should be "carrying" a pointer to the
            // correct TLBEntry.
            new_entry = sender_state->tlbEntry;
            assert(new_entry);
            local_entry = new_entry;

            if (allocationPolicy) {
                DPRINTF(GPUTLB, "allocating entry w/ addr %#x\n",
                        virt_page_addr);

                local_entry = insert(virt_page_addr, *new_entry);
            }

            assert(local_entry);
        }

        /**
         * At this point the packet carries an up-to-date tlbEntry pointer
         * in its senderState.
         * Next step is to do the paging protection checks.
         */
        DPRINTF(GPUTLB, "Entry found with vaddr %#x,  doing protection checks "
                "while paddr was %#x.\n", local_entry->vaddr,
                local_entry->paddr);

        pagingProtectionChecks(tc, pkt, local_entry, mode);
        int page_size = local_entry->size();
        Addr paddr = local_entry->paddr | (vaddr & (page_size - 1));
        DPRINTF(GPUTLB, "Translated %#x -> %#x.\n", vaddr, paddr);

        // Since this packet will be sent through the cpu side slave port,
        // it must be converted to a response pkt if it is not one already
        if (pkt->isRequest()) {
            pkt->makeTimingResponse();
        }

        pkt->req->setPaddr(paddr);

        if (local_entry->uncacheable) {
             pkt->req->setFlags(Request::UNCACHEABLE);
        }

        //send packet back to coalescer
        cpuSidePort[0]->sendTimingResp(pkt);
        //schedule cleanup event
        cleanupQueue.push(virt_page_addr);

        // schedule this only once per cycle.
        // The check is required because we might have multiple translations
        // returning the same cycle
        // this is a maximum priority event and must be on the same cycle
        // as the cleanup event in TLBCoalescer to avoid a race with
        // IssueProbeEvent caused by TLBCoalescer::MemSidePort::recvReqRetry
        if (!cleanupEvent.scheduled())
            schedule(cleanupEvent, curTick());
    }

    /**
     * Here we take the appropriate actions based on the result of the
     * TLB lookup.
     */
    void
    GpuTLB::translationReturn(Addr virtPageAddr, tlbOutcome outcome,
                              PacketPtr pkt)
    {
        DPRINTF(GPUTLB, "Triggered TLBEvent for addr %#x\n", virtPageAddr);

        assert(translationReturnEvent[virtPageAddr]);
        assert(pkt);

        TranslationState *tmp_sender_state =
            safe_cast<TranslationState*>(pkt->senderState);

        int req_cnt = tmp_sender_state->reqCnt.back();
        bool update_stats = !tmp_sender_state->prefetch;


        if (outcome == TLB_HIT) {
            handleTranslationReturn(virtPageAddr, TLB_HIT, pkt);

            if (update_stats) {
                accessCycles += (req_cnt * curTick());
                localCycles += curTick();
            }

        } else if (outcome == TLB_MISS) {

            DPRINTF(GPUTLB, "This is a TLB miss\n");
            if (update_stats) {
                accessCycles += (req_cnt*curTick());
                localCycles += curTick();
            }

            if (hasMemSidePort) {
                // the one cyle added here represent the delay from when we get
                // the reply back till when we propagate it to the coalescer
                // above.
                if (update_stats) {
                    accessCycles += (req_cnt * 1);
                    localCycles += 1;
                }

                /**
                 * There is a TLB below. Send the coalesced request.
                 * We actually send the very first packet of all the
                 * pending packets for this virtual page address.
                 */
                if (!memSidePort[0]->sendTimingReq(pkt)) {
                    DPRINTF(GPUTLB, "Failed sending translation request to "
                            "lower level TLB for addr %#x\n", virtPageAddr);

                    memSidePort[0]->retries.push_back(pkt);
                } else {
                    DPRINTF(GPUTLB, "Sent translation request to lower level "
                            "TLB for addr %#x\n", virtPageAddr);
                }
            } else {
                //this is the last level TLB. Start a page walk
                DPRINTF(GPUTLB, "Last level TLB - start a page walk for "
                        "addr %#x\n", virtPageAddr);

                if (update_stats)
                    pageTableCycles -= (req_cnt*curTick());

                TLBEvent *tlb_event = translationReturnEvent[virtPageAddr];
                assert(tlb_event);
                tlb_event->updateOutcome(PAGE_WALK);
                schedule(tlb_event, curTick() + ticks(missLatency2));
            }
        } else if (outcome == PAGE_WALK) {
            if (update_stats)
                pageTableCycles += (req_cnt*curTick());

            // Need to access the page table and update the TLB
            DPRINTF(GPUTLB, "Doing a page walk for address %#x\n",
                    virtPageAddr);

            TranslationState *sender_state =
                safe_cast<TranslationState*>(pkt->senderState);

            Process *p = sender_state->tc->getProcessPtr();
            Addr vaddr = pkt->req->getVaddr();
    #ifndef NDEBUG
            Addr alignedVaddr = p->pTable->pageAlign(vaddr);
            assert(alignedVaddr == virtPageAddr);
    #endif
            const EmulationPageTable::Entry *pte = p->pTable->lookup(vaddr);
            if (!pte && sender_state->tlbMode != BaseTLB::Execute &&
                    p->fixupStackFault(vaddr)) {
                pte = p->pTable->lookup(vaddr);
            }

            if (pte) {
                DPRINTF(GPUTLB, "Mapping %#x to %#x\n", alignedVaddr,
                        pte->paddr);

                sender_state->tlbEntry =
                    new GpuTlbEntry(0, virtPageAddr, pte->paddr, true);
            } else {
                sender_state->tlbEntry =
                    new GpuTlbEntry(0, 0, 0, false);
            }

            handleTranslationReturn(virtPageAddr, TLB_MISS, pkt);
        } else if (outcome == MISS_RETURN) {
            /** we add an extra cycle in the return path of the translation
             * requests in between the various TLB levels.
             */
            handleTranslationReturn(virtPageAddr, TLB_MISS, pkt);
        } else {
            assert(false);
        }
    }

    void
    GpuTLB::TLBEvent::process()
    {
        tlb->translationReturn(virtPageAddr, outcome, pkt);
    }

    const char*
    GpuTLB::TLBEvent::description() const
    {
        return "trigger translationDoneEvent";
    }

    void
    GpuTLB::TLBEvent::updateOutcome(tlbOutcome _outcome)
    {
        outcome = _outcome;
    }

    Addr
    GpuTLB::TLBEvent::getTLBEventVaddr()
    {
        return virtPageAddr;
    }

    /*
     * recvTiming receives a coalesced timing request from a TLBCoalescer
     * and it calls issueTLBLookup()
     * It only rejects the packet if we have exceeded the max
     * outstanding number of requests for the TLB
     */
    bool
    GpuTLB::CpuSidePort::recvTimingReq(PacketPtr pkt)
    {
        if (tlb->outstandingReqs < tlb->maxCoalescedReqs) {
            tlb->issueTLBLookup(pkt);
            // update number of outstanding translation requests
            tlb->outstandingReqs++;
            return true;
         } else {
            DPRINTF(GPUTLB, "Reached maxCoalescedReqs number %d\n",
                    tlb->outstandingReqs);
            return false;
         }
    }

    /**
     * handleFuncTranslationReturn is called on a TLB hit,
     * when a TLB miss returns or when a page fault returns.
     * It updates LRU, inserts the TLB entry on a miss
     * depending on the allocation policy and does the required
     * protection checks. It does NOT create a new packet to
     * update the packet's addr; this is done in hsail-gpu code.
     */
    void
    GpuTLB::handleFuncTranslationReturn(PacketPtr pkt, tlbOutcome tlb_outcome)
    {
        TranslationState *sender_state =
            safe_cast<TranslationState*>(pkt->senderState);

        ThreadContext *tc = sender_state->tc;
        Mode mode = sender_state->tlbMode;
        Addr vaddr = pkt->req->getVaddr();

        GpuTlbEntry *local_entry, *new_entry;

        if (tlb_outcome == TLB_HIT) {
            DPRINTF(GPUTLB, "Functional Translation Done - TLB hit for addr "
                    "%#x\n", vaddr);

            local_entry = sender_state->tlbEntry;
        } else {
            DPRINTF(GPUTLB, "Functional Translation Done - TLB miss for addr "
                    "%#x\n", vaddr);

            // We are returning either from a page walk or from a hit at a lower
            // TLB level. The senderState should be "carrying" a pointer to the
            // correct TLBEntry.
            new_entry = sender_state->tlbEntry;
            assert(new_entry);
            local_entry = new_entry;

            if (allocationPolicy) {
                Addr virt_page_addr = roundDown(vaddr, TheISA::PageBytes);

                DPRINTF(GPUTLB, "allocating entry w/ addr %#x\n",
                        virt_page_addr);

                local_entry = insert(virt_page_addr, *new_entry);
            }

            assert(local_entry);
        }

        DPRINTF(GPUTLB, "Entry found with vaddr %#x, doing protection checks "
                "while paddr was %#x.\n", local_entry->vaddr,
                local_entry->paddr);

        // Do paging checks if it's a normal functional access.  If it's for a
        // prefetch, then sometimes you can try to prefetch something that won't
        // pass protection. We don't actually want to fault becuase there is no
        // demand access to deem this a violation.  Just put it in the TLB and
        // it will fault if indeed a future demand access touches it in
        // violation.
        if (!sender_state->prefetch && sender_state->tlbEntry->valid)
            pagingProtectionChecks(tc, pkt, local_entry, mode);

        int page_size = local_entry->size();
        Addr paddr = local_entry->paddr | (vaddr & (page_size - 1));
        DPRINTF(GPUTLB, "Translated %#x -> %#x.\n", vaddr, paddr);

        pkt->req->setPaddr(paddr);

        if (local_entry->uncacheable)
             pkt->req->setFlags(Request::UNCACHEABLE);
    }

    // This is used for atomic translations. Need to
    // make it all happen during the same cycle.
    void
    GpuTLB::CpuSidePort::recvFunctional(PacketPtr pkt)
    {
        TranslationState *sender_state =
            safe_cast<TranslationState*>(pkt->senderState);

        ThreadContext *tc = sender_state->tc;
        bool update_stats = !sender_state->prefetch;

        Addr virt_page_addr = roundDown(pkt->req->getVaddr(),
                                        TheISA::PageBytes);

        if (update_stats)
            tlb->updatePageFootprint(virt_page_addr);

        // do the TLB lookup without updating the stats
        bool success = tlb->tlbLookup(pkt->req, tc, update_stats);
        tlbOutcome tlb_outcome = success ? TLB_HIT : TLB_MISS;

        // functional mode means no coalescing
        // global metrics are the same as the local metrics
        if (update_stats) {
            tlb->globalNumTLBAccesses++;

            if (success) {
                sender_state->hitLevel = sender_state->reqCnt.size();
                tlb->globalNumTLBHits++;
            }
        }

        if (!success) {
            if (update_stats)
                tlb->globalNumTLBMisses++;
            if (tlb->hasMemSidePort) {
                // there is a TLB below -> propagate down the TLB hierarchy
                tlb->memSidePort[0]->sendFunctional(pkt);
                // If no valid translation from a prefetch, then just return
                if (sender_state->prefetch && !pkt->req->hasPaddr())
                    return;
            } else {
                // Need to access the page table and update the TLB
                DPRINTF(GPUTLB, "Doing a page walk for address %#x\n",
                        virt_page_addr);

                Process *p = tc->getProcessPtr();

                Addr vaddr = pkt->req->getVaddr();
    #ifndef NDEBUG
                Addr alignedVaddr = p->pTable->pageAlign(vaddr);
                assert(alignedVaddr == virt_page_addr);
    #endif

                const EmulationPageTable::Entry *pte =
                        p->pTable->lookup(vaddr);
                if (!pte && sender_state->tlbMode != BaseTLB::Execute &&
                        p->fixupStackFault(vaddr)) {
                    pte = p->pTable->lookup(vaddr);
                }

                if (!sender_state->prefetch) {
                    // no PageFaults are permitted after
                    // the second page table lookup
                    assert(success);

                    DPRINTF(GPUTLB, "Mapping %#x to %#x\n", alignedVaddr,
                            pte->paddr);

                    sender_state->tlbEntry =
                        new GpuTlbEntry(0, virt_page_addr,
                                        pte->paddr, success);
                } else {
                    // If this was a prefetch, then do the normal thing if it
                    // was a successful translation.  Otherwise, send an empty
                    // TLB entry back so that it can be figured out as empty and
                    // handled accordingly.
                    if (pte) {
                        DPRINTF(GPUTLB, "Mapping %#x to %#x\n", alignedVaddr,
                                pte->paddr);

                        sender_state->tlbEntry =
                            new GpuTlbEntry(0, virt_page_addr,
                                            pte->paddr, success);
                    } else {
                        DPRINTF(GPUPrefetch, "Prefetch failed %#x\n",
                                alignedVaddr);

                        sender_state->tlbEntry = new GpuTlbEntry();

                        return;
                    }
                }
            }
        } else {
            DPRINTF(GPUPrefetch, "Functional Hit for vaddr %#x\n",
                    tlb->lookup(pkt->req->getVaddr()));

            GpuTlbEntry *entry = tlb->lookup(pkt->req->getVaddr(),
                                             update_stats);

            assert(entry);

            sender_state->tlbEntry =
                new GpuTlbEntry(0, entry->vaddr, entry->paddr, entry->valid);
        }
        // This is the function that would populate pkt->req with the paddr of
        // the translation. But if no translation happens (i.e Prefetch fails)
        // then the early returns in the above code wiill keep this function
        // from executing.
        tlb->handleFuncTranslationReturn(pkt, tlb_outcome);
    }

    void
    GpuTLB::CpuSidePort::recvReqRetry()
    {
        // The CPUSidePort never sends anything but replies. No retries
        // expected.
        assert(false);
    }

    AddrRangeList
    GpuTLB::CpuSidePort::getAddrRanges() const
    {
        // currently not checked by the master
        AddrRangeList ranges;

        return ranges;
    }

    /**
     * MemSidePort receives the packet back.
     * We need to call the handleTranslationReturn
     * and propagate up the hierarchy.
     */
    bool
    GpuTLB::MemSidePort::recvTimingResp(PacketPtr pkt)
    {
        Addr virt_page_addr = roundDown(pkt->req->getVaddr(),
                                        TheISA::PageBytes);

        DPRINTF(GPUTLB, "MemSidePort recvTiming for virt_page_addr %#x\n",
                virt_page_addr);

        TLBEvent *tlb_event = tlb->translationReturnEvent[virt_page_addr];
        assert(tlb_event);
        assert(virt_page_addr == tlb_event->getTLBEventVaddr());

        tlb_event->updateOutcome(MISS_RETURN);
        tlb->schedule(tlb_event, curTick()+tlb->ticks(1));

        return true;
    }

    void
    GpuTLB::MemSidePort::recvReqRetry()
    {
        // No retries should reach the TLB. The retries
        // should only reach the TLBCoalescer.
        assert(false);
    }

    void
    GpuTLB::cleanup()
    {
        while (!cleanupQueue.empty()) {
            Addr cleanup_addr = cleanupQueue.front();
            cleanupQueue.pop();

            // delete TLBEvent
            TLBEvent * old_tlb_event = translationReturnEvent[cleanup_addr];
            delete old_tlb_event;
            translationReturnEvent.erase(cleanup_addr);

            // update number of outstanding requests
            outstandingReqs--;
        }

        /** the higher level coalescer should retry if it has
         * any pending requests.
         */
        for (int i = 0; i < cpuSidePort.size(); ++i) {
            cpuSidePort[i]->sendRetryReq();
        }
    }

    void
    GpuTLB::updatePageFootprint(Addr virt_page_addr)
    {

        std::pair<AccessPatternTable::iterator, bool> ret;

        AccessInfo tmp_access_info;
        tmp_access_info.lastTimeAccessed = 0;
        tmp_access_info.accessesPerPage = 0;
        tmp_access_info.totalReuseDistance = 0;
        tmp_access_info.sumDistance = 0;
        tmp_access_info.meanDistance = 0;

        ret = TLBFootprint.insert(AccessPatternTable::value_type(virt_page_addr,
                                  tmp_access_info));

        bool first_page_access = ret.second;

        if (first_page_access) {
            numUniquePages++;
        } else  {
            int accessed_before;
            accessed_before  = curTick() - ret.first->second.lastTimeAccessed;
            ret.first->second.totalReuseDistance += accessed_before;
        }

        ret.first->second.accessesPerPage++;
        ret.first->second.lastTimeAccessed = curTick();

        if (accessDistance) {
            ret.first->second.localTLBAccesses
                .push_back(localNumTLBAccesses.value());
        }
    }

    void
    GpuTLB::exitCallback()
    {
        std::ostream *page_stat_file = nullptr;

        if (accessDistance) {

            // print per page statistics to a separate file (.csv format)
            // simout is the gem5 output directory (default is m5out or the one
            // specified with -d
            page_stat_file = simout.create(name().c_str())->stream();

            // print header
            *page_stat_file << "page,max_access_distance,mean_access_distance, "
                            << "stddev_distance" << std::endl;
        }

        // update avg. reuse distance footprint
        AccessPatternTable::iterator iter, iter_begin, iter_end;
        unsigned int sum_avg_reuse_distance_per_page = 0;

        // iterate through all pages seen by this TLB
        for (iter = TLBFootprint.begin(); iter != TLBFootprint.end(); iter++) {
            sum_avg_reuse_distance_per_page += iter->second.totalReuseDistance /
                                               iter->second.accessesPerPage;

            if (accessDistance) {
                unsigned int tmp = iter->second.localTLBAccesses[0];
                unsigned int prev = tmp;

                for (int i = 0; i < iter->second.localTLBAccesses.size(); ++i) {
                    if (i) {
                        tmp = prev + 1;
                    }

                    prev = iter->second.localTLBAccesses[i];
                    // update the localTLBAccesses value
                    // with the actual differece
                    iter->second.localTLBAccesses[i] -= tmp;
                    // compute the sum of AccessDistance per page
                    // used later for mean
                    iter->second.sumDistance +=
                        iter->second.localTLBAccesses[i];
                }

                iter->second.meanDistance =
                    iter->second.sumDistance / iter->second.accessesPerPage;

                // compute std_dev and max  (we need a second round because we
                // need to know the mean value
                unsigned int max_distance = 0;
                unsigned int stddev_distance = 0;

                for (int i = 0; i < iter->second.localTLBAccesses.size(); ++i) {
                    unsigned int tmp_access_distance =
                        iter->second.localTLBAccesses[i];

                    if (tmp_access_distance > max_distance) {
                        max_distance = tmp_access_distance;
                    }

                    unsigned int diff =
                        tmp_access_distance - iter->second.meanDistance;
                    stddev_distance += pow(diff, 2);

                }

                stddev_distance =
                    sqrt(stddev_distance/iter->second.accessesPerPage);

                if (page_stat_file) {
                    *page_stat_file << std::hex << iter->first << ",";
                    *page_stat_file << std::dec << max_distance << ",";
                    *page_stat_file << std::dec << iter->second.meanDistance
                                    << ",";
                    *page_stat_file << std::dec << stddev_distance;
                    *page_stat_file << std::endl;
                }

                // erase the localTLBAccesses array
                iter->second.localTLBAccesses.clear();
            }
        }

        if (!TLBFootprint.empty()) {
            avgReuseDistance =
                sum_avg_reuse_distance_per_page / TLBFootprint.size();
        }

        //clear the TLBFootprint map
        TLBFootprint.clear();
    }
} // namespace X86ISA

X86ISA::GpuTLB*
X86GPUTLBParams::create()
{
    return new X86ISA::GpuTLB(this);
}