summaryrefslogtreecommitdiff
path: root/src/mem/dram_ctrl.cc
blob: 2b053d78e8c990fabe3d3b637425c81506174c00 (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
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
/*
 * Copyright (c) 2010-2014 ARM Limited
 * All rights reserved
 *
 * The license below extends only to copyright in the software and shall
 * not be construed as granting a license to any other intellectual
 * property including but not limited to intellectual property relating
 * to a hardware implementation of the functionality of the software
 * licensed hereunder.  You may use the software subject to the license
 * terms below provided that you ensure that this notice is replicated
 * unmodified and in its entirety in all distributions of the software,
 * modified or unmodified, in source code or in binary form.
 *
 * Copyright (c) 2013 Amin Farmahini-Farahani
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are
 * met: redistributions of source code must retain the above copyright
 * notice, this list of conditions and the following disclaimer;
 * 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;
 * neither the name of the copyright holders 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
 * OWNER 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.
 *
 * Authors: Andreas Hansson
 *          Ani Udipi
 *          Neha Agarwal
 */

#include "base/bitfield.hh"
#include "base/trace.hh"
#include "debug/DRAM.hh"
#include "debug/DRAMPower.hh"
#include "debug/DRAMState.hh"
#include "debug/Drain.hh"
#include "mem/dram_ctrl.hh"
#include "sim/system.hh"

using namespace std;

DRAMCtrl::DRAMCtrl(const DRAMCtrlParams* p) :
    AbstractMemory(p),
    port(name() + ".port", *this),
    retryRdReq(false), retryWrReq(false),
    busState(READ),
    nextReqEvent(this), respondEvent(this), activateEvent(this),
    prechargeEvent(this), refreshEvent(this), powerEvent(this),
    drainManager(NULL),
    deviceBusWidth(p->device_bus_width), burstLength(p->burst_length),
    deviceRowBufferSize(p->device_rowbuffer_size),
    devicesPerRank(p->devices_per_rank),
    burstSize((devicesPerRank * burstLength * deviceBusWidth) / 8),
    rowBufferSize(devicesPerRank * deviceRowBufferSize),
    columnsPerRowBuffer(rowBufferSize / burstSize),
    ranksPerChannel(p->ranks_per_channel),
    banksPerRank(p->banks_per_rank), channels(p->channels), rowsPerBank(0),
    readBufferSize(p->read_buffer_size),
    writeBufferSize(p->write_buffer_size),
    writeHighThreshold(writeBufferSize * p->write_high_thresh_perc / 100.0),
    writeLowThreshold(writeBufferSize * p->write_low_thresh_perc / 100.0),
    minWritesPerSwitch(p->min_writes_per_switch),
    writesThisTime(0), readsThisTime(0),
    tCK(p->tCK), tWTR(p->tWTR), tRTW(p->tRTW), tBURST(p->tBURST),
    tRCD(p->tRCD), tCL(p->tCL), tRP(p->tRP), tRAS(p->tRAS), tWR(p->tWR),
    tRTP(p->tRTP), tRFC(p->tRFC), tREFI(p->tREFI), tRRD(p->tRRD),
    tXAW(p->tXAW), activationLimit(p->activation_limit),
    memSchedPolicy(p->mem_sched_policy), addrMapping(p->addr_mapping),
    pageMgmt(p->page_policy),
    maxAccessesPerRow(p->max_accesses_per_row),
    frontendLatency(p->static_frontend_latency),
    backendLatency(p->static_backend_latency),
    busBusyUntil(0), refreshDueAt(0), refreshState(REF_IDLE),
    pwrStateTrans(PWR_IDLE), pwrState(PWR_IDLE), prevArrival(0),
    nextReqTime(0), pwrStateTick(0), numBanksActive(0)
{
    // create the bank states based on the dimensions of the ranks and
    // banks
    banks.resize(ranksPerChannel);
    actTicks.resize(ranksPerChannel);
    for (size_t c = 0; c < ranksPerChannel; ++c) {
        banks[c].resize(banksPerRank);
        actTicks[c].resize(activationLimit, 0);
    }

    // set the bank indices
    for (int r = 0; r < ranksPerChannel; r++) {
        for (int b = 0; b < banksPerRank; b++) {
            banks[r][b].rank = r;
            banks[r][b].bank = b;
        }
    }

    // perform a basic check of the write thresholds
    if (p->write_low_thresh_perc >= p->write_high_thresh_perc)
        fatal("Write buffer low threshold %d must be smaller than the "
              "high threshold %d\n", p->write_low_thresh_perc,
              p->write_high_thresh_perc);

    // determine the rows per bank by looking at the total capacity
    uint64_t capacity = ULL(1) << ceilLog2(AbstractMemory::size());

    DPRINTF(DRAM, "Memory capacity %lld (%lld) bytes\n", capacity,
            AbstractMemory::size());

    DPRINTF(DRAM, "Row buffer size %d bytes with %d columns per row buffer\n",
            rowBufferSize, columnsPerRowBuffer);

    rowsPerBank = capacity / (rowBufferSize * banksPerRank * ranksPerChannel);

    if (range.interleaved()) {
        if (channels != range.stripes())
            fatal("%s has %d interleaved address stripes but %d channel(s)\n",
                  name(), range.stripes(), channels);

        if (addrMapping == Enums::RoRaBaChCo) {
            if (rowBufferSize != range.granularity()) {
                fatal("Interleaving of %s doesn't match RoRaBaChCo "
                      "address map\n", name());
            }
        } else if (addrMapping == Enums::RoRaBaCoCh) {
            if (system()->cacheLineSize() != range.granularity()) {
                fatal("Interleaving of %s doesn't match RoRaBaCoCh "
                      "address map\n", name());
            }
        } else if (addrMapping == Enums::RoCoRaBaCh) {
            if (system()->cacheLineSize() != range.granularity())
                fatal("Interleaving of %s doesn't match RoCoRaBaCh "
                      "address map\n", name());
        }
    }

    // some basic sanity checks
    if (tREFI <= tRP || tREFI <= tRFC) {
        fatal("tREFI (%d) must be larger than tRP (%d) and tRFC (%d)\n",
              tREFI, tRP, tRFC);
    }
}

void
DRAMCtrl::init()
{
    if (!port.isConnected()) {
        fatal("DRAMCtrl %s is unconnected!\n", name());
    } else {
        port.sendRangeChange();
    }
}

void
DRAMCtrl::startup()
{
    // update the start tick for the precharge accounting to the
    // current tick
    pwrStateTick = curTick();

    // shift the bus busy time sufficiently far ahead that we never
    // have to worry about negative values when computing the time for
    // the next request, this will add an insignificant bubble at the
    // start of simulation
    busBusyUntil = curTick() + tRP + tRCD + tCL;

    // kick off the refresh, and give ourselves enough time to
    // precharge
    schedule(refreshEvent, curTick() + tREFI - tRP);
}

Tick
DRAMCtrl::recvAtomic(PacketPtr pkt)
{
    DPRINTF(DRAM, "recvAtomic: %s 0x%x\n", pkt->cmdString(), pkt->getAddr());

    // do the actual memory access and turn the packet into a response
    access(pkt);

    Tick latency = 0;
    if (!pkt->memInhibitAsserted() && pkt->hasData()) {
        // this value is not supposed to be accurate, just enough to
        // keep things going, mimic a closed page
        latency = tRP + tRCD + tCL;
    }
    return latency;
}

bool
DRAMCtrl::readQueueFull(unsigned int neededEntries) const
{
    DPRINTF(DRAM, "Read queue limit %d, current size %d, entries needed %d\n",
            readBufferSize, readQueue.size() + respQueue.size(),
            neededEntries);

    return
        (readQueue.size() + respQueue.size() + neededEntries) > readBufferSize;
}

bool
DRAMCtrl::writeQueueFull(unsigned int neededEntries) const
{
    DPRINTF(DRAM, "Write queue limit %d, current size %d, entries needed %d\n",
            writeBufferSize, writeQueue.size(), neededEntries);
    return (writeQueue.size() + neededEntries) > writeBufferSize;
}

DRAMCtrl::DRAMPacket*
DRAMCtrl::decodeAddr(PacketPtr pkt, Addr dramPktAddr, unsigned size,
                       bool isRead)
{
    // decode the address based on the address mapping scheme, with
    // Ro, Ra, Co, Ba and Ch denoting row, rank, column, bank and
    // channel, respectively
    uint8_t rank;
    uint8_t bank;
    // use a 64-bit unsigned during the computations as the row is
    // always the top bits, and check before creating the DRAMPacket
    uint64_t row;

    // truncate the address to the access granularity
    Addr addr = dramPktAddr / burstSize;

    // we have removed the lowest order address bits that denote the
    // position within the column
    if (addrMapping == Enums::RoRaBaChCo) {
        // the lowest order bits denote the column to ensure that
        // sequential cache lines occupy the same row
        addr = addr / columnsPerRowBuffer;

        // take out the channel part of the address
        addr = addr / channels;

        // after the channel bits, get the bank bits to interleave
        // over the banks
        bank = addr % banksPerRank;
        addr = addr / banksPerRank;

        // after the bank, we get the rank bits which thus interleaves
        // over the ranks
        rank = addr % ranksPerChannel;
        addr = addr / ranksPerChannel;

        // lastly, get the row bits
        row = addr % rowsPerBank;
        addr = addr / rowsPerBank;
    } else if (addrMapping == Enums::RoRaBaCoCh) {
        // take out the channel part of the address
        addr = addr / channels;

        // next, the column
        addr = addr / columnsPerRowBuffer;

        // after the column bits, we get the bank bits to interleave
        // over the banks
        bank = addr % banksPerRank;
        addr = addr / banksPerRank;

        // after the bank, we get the rank bits which thus interleaves
        // over the ranks
        rank = addr % ranksPerChannel;
        addr = addr / ranksPerChannel;

        // lastly, get the row bits
        row = addr % rowsPerBank;
        addr = addr / rowsPerBank;
    } else if (addrMapping == Enums::RoCoRaBaCh) {
        // optimise for closed page mode and utilise maximum
        // parallelism of the DRAM (at the cost of power)

        // take out the channel part of the address, not that this has
        // to match with how accesses are interleaved between the
        // controllers in the address mapping
        addr = addr / channels;

        // start with the bank bits, as this provides the maximum
        // opportunity for parallelism between requests
        bank = addr % banksPerRank;
        addr = addr / banksPerRank;

        // next get the rank bits
        rank = addr % ranksPerChannel;
        addr = addr / ranksPerChannel;

        // next the column bits which we do not need to keep track of
        // and simply skip past
        addr = addr / columnsPerRowBuffer;

        // lastly, get the row bits
        row = addr % rowsPerBank;
        addr = addr / rowsPerBank;
    } else
        panic("Unknown address mapping policy chosen!");

    assert(rank < ranksPerChannel);
    assert(bank < banksPerRank);
    assert(row < rowsPerBank);
    assert(row < Bank::NO_ROW);

    DPRINTF(DRAM, "Address: %lld Rank %d Bank %d Row %d\n",
            dramPktAddr, rank, bank, row);

    // create the corresponding DRAM packet with the entry time and
    // ready time set to the current tick, the latter will be updated
    // later
    uint16_t bank_id = banksPerRank * rank + bank;
    return new DRAMPacket(pkt, isRead, rank, bank, row, bank_id, dramPktAddr,
                          size, banks[rank][bank]);
}

void
DRAMCtrl::addToReadQueue(PacketPtr pkt, unsigned int pktCount)
{
    // only add to the read queue here. whenever the request is
    // eventually done, set the readyTime, and call schedule()
    assert(!pkt->isWrite());

    assert(pktCount != 0);

    // if the request size is larger than burst size, the pkt is split into
    // multiple DRAM packets
    // Note if the pkt starting address is not aligened to burst size, the
    // address of first DRAM packet is kept unaliged. Subsequent DRAM packets
    // are aligned to burst size boundaries. This is to ensure we accurately
    // check read packets against packets in write queue.
    Addr addr = pkt->getAddr();
    unsigned pktsServicedByWrQ = 0;
    BurstHelper* burst_helper = NULL;
    for (int cnt = 0; cnt < pktCount; ++cnt) {
        unsigned size = std::min((addr | (burstSize - 1)) + 1,
                        pkt->getAddr() + pkt->getSize()) - addr;
        readPktSize[ceilLog2(size)]++;
        readBursts++;

        // First check write buffer to see if the data is already at
        // the controller
        bool foundInWrQ = false;
        for (auto i = writeQueue.begin(); i != writeQueue.end(); ++i) {
            // check if the read is subsumed in the write entry we are
            // looking at
            if ((*i)->addr <= addr &&
                (addr + size) <= ((*i)->addr + (*i)->size)) {
                foundInWrQ = true;
                servicedByWrQ++;
                pktsServicedByWrQ++;
                DPRINTF(DRAM, "Read to addr %lld with size %d serviced by "
                        "write queue\n", addr, size);
                bytesReadWrQ += burstSize;
                break;
            }
        }

        // If not found in the write q, make a DRAM packet and
        // push it onto the read queue
        if (!foundInWrQ) {

            // Make the burst helper for split packets
            if (pktCount > 1 && burst_helper == NULL) {
                DPRINTF(DRAM, "Read to addr %lld translates to %d "
                        "dram requests\n", pkt->getAddr(), pktCount);
                burst_helper = new BurstHelper(pktCount);
            }

            DRAMPacket* dram_pkt = decodeAddr(pkt, addr, size, true);
            dram_pkt->burstHelper = burst_helper;

            assert(!readQueueFull(1));
            rdQLenPdf[readQueue.size() + respQueue.size()]++;

            DPRINTF(DRAM, "Adding to read queue\n");

            readQueue.push_back(dram_pkt);

            // Update stats
            avgRdQLen = readQueue.size() + respQueue.size();
        }

        // Starting address of next dram pkt (aligend to burstSize boundary)
        addr = (addr | (burstSize - 1)) + 1;
    }

    // If all packets are serviced by write queue, we send the repsonse back
    if (pktsServicedByWrQ == pktCount) {
        accessAndRespond(pkt, frontendLatency);
        return;
    }

    // Update how many split packets are serviced by write queue
    if (burst_helper != NULL)
        burst_helper->burstsServiced = pktsServicedByWrQ;

    // If we are not already scheduled to get a request out of the
    // queue, do so now
    if (!nextReqEvent.scheduled()) {
        DPRINTF(DRAM, "Request scheduled immediately\n");
        schedule(nextReqEvent, curTick());
    }
}

void
DRAMCtrl::addToWriteQueue(PacketPtr pkt, unsigned int pktCount)
{
    // only add to the write queue here. whenever the request is
    // eventually done, set the readyTime, and call schedule()
    assert(pkt->isWrite());

    // if the request size is larger than burst size, the pkt is split into
    // multiple DRAM packets
    Addr addr = pkt->getAddr();
    for (int cnt = 0; cnt < pktCount; ++cnt) {
        unsigned size = std::min((addr | (burstSize - 1)) + 1,
                        pkt->getAddr() + pkt->getSize()) - addr;
        writePktSize[ceilLog2(size)]++;
        writeBursts++;

        // see if we can merge with an existing item in the write
        // queue and keep track of whether we have merged or not so we
        // can stop at that point and also avoid enqueueing a new
        // request
        bool merged = false;
        auto w = writeQueue.begin();

        while(!merged && w != writeQueue.end()) {
            // either of the two could be first, if they are the same
            // it does not matter which way we go
            if ((*w)->addr >= addr) {
                // the existing one starts after the new one, figure
                // out where the new one ends with respect to the
                // existing one
                if ((addr + size) >= ((*w)->addr + (*w)->size)) {
                    // check if the existing one is completely
                    // subsumed in the new one
                    DPRINTF(DRAM, "Merging write covering existing burst\n");
                    merged = true;
                    // update both the address and the size
                    (*w)->addr = addr;
                    (*w)->size = size;
                } else if ((addr + size) >= (*w)->addr &&
                           ((*w)->addr + (*w)->size - addr) <= burstSize) {
                    // the new one is just before or partially
                    // overlapping with the existing one, and together
                    // they fit within a burst
                    DPRINTF(DRAM, "Merging write before existing burst\n");
                    merged = true;
                    // the existing queue item needs to be adjusted with
                    // respect to both address and size
                    (*w)->size = (*w)->addr + (*w)->size - addr;
                    (*w)->addr = addr;
                }
            } else {
                // the new one starts after the current one, figure
                // out where the existing one ends with respect to the
                // new one
                if (((*w)->addr + (*w)->size) >= (addr + size)) {
                    // check if the new one is completely subsumed in the
                    // existing one
                    DPRINTF(DRAM, "Merging write into existing burst\n");
                    merged = true;
                    // no adjustments necessary
                } else if (((*w)->addr + (*w)->size) >= addr &&
                           (addr + size - (*w)->addr) <= burstSize) {
                    // the existing one is just before or partially
                    // overlapping with the new one, and together
                    // they fit within a burst
                    DPRINTF(DRAM, "Merging write after existing burst\n");
                    merged = true;
                    // the address is right, and only the size has
                    // to be adjusted
                    (*w)->size = addr + size - (*w)->addr;
                }
            }
            ++w;
        }

        // if the item was not merged we need to create a new write
        // and enqueue it
        if (!merged) {
            DRAMPacket* dram_pkt = decodeAddr(pkt, addr, size, false);

            assert(writeQueue.size() < writeBufferSize);
            wrQLenPdf[writeQueue.size()]++;

            DPRINTF(DRAM, "Adding to write queue\n");

            writeQueue.push_back(dram_pkt);

            // Update stats
            avgWrQLen = writeQueue.size();
        } else {
            // keep track of the fact that this burst effectively
            // disappeared as it was merged with an existing one
            mergedWrBursts++;
        }

        // Starting address of next dram pkt (aligend to burstSize boundary)
        addr = (addr | (burstSize - 1)) + 1;
    }

    // we do not wait for the writes to be send to the actual memory,
    // but instead take responsibility for the consistency here and
    // snoop the write queue for any upcoming reads
    // @todo, if a pkt size is larger than burst size, we might need a
    // different front end latency
    accessAndRespond(pkt, frontendLatency);

    // If we are not already scheduled to get a request out of the
    // queue, do so now
    if (!nextReqEvent.scheduled()) {
        DPRINTF(DRAM, "Request scheduled immediately\n");
        schedule(nextReqEvent, curTick());
    }
}

void
DRAMCtrl::printQs() const {
    DPRINTF(DRAM, "===READ QUEUE===\n\n");
    for (auto i = readQueue.begin() ;  i != readQueue.end() ; ++i) {
        DPRINTF(DRAM, "Read %lu\n", (*i)->addr);
    }
    DPRINTF(DRAM, "\n===RESP QUEUE===\n\n");
    for (auto i = respQueue.begin() ;  i != respQueue.end() ; ++i) {
        DPRINTF(DRAM, "Response %lu\n", (*i)->addr);
    }
    DPRINTF(DRAM, "\n===WRITE QUEUE===\n\n");
    for (auto i = writeQueue.begin() ;  i != writeQueue.end() ; ++i) {
        DPRINTF(DRAM, "Write %lu\n", (*i)->addr);
    }
}

bool
DRAMCtrl::recvTimingReq(PacketPtr pkt)
{
    /// @todo temporary hack to deal with memory corruption issues until
    /// 4-phase transactions are complete
    for (int x = 0; x < pendingDelete.size(); x++)
        delete pendingDelete[x];
    pendingDelete.clear();

    // This is where we enter from the outside world
    DPRINTF(DRAM, "recvTimingReq: request %s addr %lld size %d\n",
            pkt->cmdString(), pkt->getAddr(), pkt->getSize());

    // simply drop inhibited packets for now
    if (pkt->memInhibitAsserted()) {
        DPRINTF(DRAM, "Inhibited packet -- Dropping it now\n");
        pendingDelete.push_back(pkt);
        return true;
    }

    // Calc avg gap between requests
    if (prevArrival != 0) {
        totGap += curTick() - prevArrival;
    }
    prevArrival = curTick();


    // Find out how many dram packets a pkt translates to
    // If the burst size is equal or larger than the pkt size, then a pkt
    // translates to only one dram packet. Otherwise, a pkt translates to
    // multiple dram packets
    unsigned size = pkt->getSize();
    unsigned offset = pkt->getAddr() & (burstSize - 1);
    unsigned int dram_pkt_count = divCeil(offset + size, burstSize);

    // check local buffers and do not accept if full
    if (pkt->isRead()) {
        assert(size != 0);
        if (readQueueFull(dram_pkt_count)) {
            DPRINTF(DRAM, "Read queue full, not accepting\n");
            // remember that we have to retry this port
            retryRdReq = true;
            numRdRetry++;
            return false;
        } else {
            addToReadQueue(pkt, dram_pkt_count);
            readReqs++;
            bytesReadSys += size;
        }
    } else if (pkt->isWrite()) {
        assert(size != 0);
        if (writeQueueFull(dram_pkt_count)) {
            DPRINTF(DRAM, "Write queue full, not accepting\n");
            // remember that we have to retry this port
            retryWrReq = true;
            numWrRetry++;
            return false;
        } else {
            addToWriteQueue(pkt, dram_pkt_count);
            writeReqs++;
            bytesWrittenSys += size;
        }
    } else {
        DPRINTF(DRAM,"Neither read nor write, ignore timing\n");
        neitherReadNorWrite++;
        accessAndRespond(pkt, 1);
    }

    return true;
}

void
DRAMCtrl::processRespondEvent()
{
    DPRINTF(DRAM,
            "processRespondEvent(): Some req has reached its readyTime\n");

    DRAMPacket* dram_pkt = respQueue.front();

    if (dram_pkt->burstHelper) {
        // it is a split packet
        dram_pkt->burstHelper->burstsServiced++;
        if (dram_pkt->burstHelper->burstsServiced ==
            dram_pkt->burstHelper->burstCount) {
            // we have now serviced all children packets of a system packet
            // so we can now respond to the requester
            // @todo we probably want to have a different front end and back
            // end latency for split packets
            accessAndRespond(dram_pkt->pkt, frontendLatency + backendLatency);
            delete dram_pkt->burstHelper;
            dram_pkt->burstHelper = NULL;
        }
    } else {
        // it is not a split packet
        accessAndRespond(dram_pkt->pkt, frontendLatency + backendLatency);
    }

    delete respQueue.front();
    respQueue.pop_front();

    if (!respQueue.empty()) {
        assert(respQueue.front()->readyTime >= curTick());
        assert(!respondEvent.scheduled());
        schedule(respondEvent, respQueue.front()->readyTime);
    } else {
        // if there is nothing left in any queue, signal a drain
        if (writeQueue.empty() && readQueue.empty() &&
            drainManager) {
            drainManager->signalDrainDone();
            drainManager = NULL;
        }
    }

    // We have made a location in the queue available at this point,
    // so if there is a read that was forced to wait, retry now
    if (retryRdReq) {
        retryRdReq = false;
        port.sendRetry();
    }
}

void
DRAMCtrl::chooseNext(std::deque<DRAMPacket*>& queue)
{
    // This method does the arbitration between requests. The chosen
    // packet is simply moved to the head of the queue. The other
    // methods know that this is the place to look. For example, with
    // FCFS, this method does nothing
    assert(!queue.empty());

    if (queue.size() == 1) {
        DPRINTF(DRAM, "Single request, nothing to do\n");
        return;
    }

    if (memSchedPolicy == Enums::fcfs) {
        // Do nothing, since the correct request is already head
    } else if (memSchedPolicy == Enums::frfcfs) {
        reorderQueue(queue);
    } else
        panic("No scheduling policy chosen\n");
}

void
DRAMCtrl::reorderQueue(std::deque<DRAMPacket*>& queue)
{
    // Only determine this when needed
    uint64_t earliest_banks = 0;

    // Search for row hits first, if no row hit is found then schedule the
    // packet to one of the earliest banks available
    bool found_earliest_pkt = false;
    auto selected_pkt_it = queue.begin();

    for (auto i = queue.begin(); i != queue.end() ; ++i) {
        DRAMPacket* dram_pkt = *i;
        const Bank& bank = dram_pkt->bankRef;
        // Check if it is a row hit
        if (bank.openRow == dram_pkt->row) {
            // FCFS within the hits
            DPRINTF(DRAM, "Row buffer hit\n");
            selected_pkt_it = i;
            break;
        } else if (!found_earliest_pkt) {
            // No row hit, go for first ready
            if (earliest_banks == 0)
                earliest_banks = minBankActAt(queue);

            // simplistic approximation of when the bank can issue an
            // activate, this is calculated in minBankActAt and could
            // be cached
            Tick act_at = bank.openRow == Bank::NO_ROW ?
                bank.actAllowedAt :
                std::max(bank.preAllowedAt, curTick()) + tRP;

            // Bank is ready or is the first available bank
            if (act_at <= curTick() ||
                bits(earliest_banks, dram_pkt->bankId, dram_pkt->bankId)) {
                // Remember the packet to be scheduled to one of the earliest
                // banks available, FCFS amongst the earliest banks
                selected_pkt_it = i;
                found_earliest_pkt = true;
            }
        }
    }

    DRAMPacket* selected_pkt = *selected_pkt_it;
    queue.erase(selected_pkt_it);
    queue.push_front(selected_pkt);
}

void
DRAMCtrl::accessAndRespond(PacketPtr pkt, Tick static_latency)
{
    DPRINTF(DRAM, "Responding to Address %lld.. ",pkt->getAddr());

    bool needsResponse = pkt->needsResponse();
    // do the actual memory access which also turns the packet into a
    // response
    access(pkt);

    // turn packet around to go back to requester if response expected
    if (needsResponse) {
        // access already turned the packet into a response
        assert(pkt->isResponse());

        // @todo someone should pay for this
        pkt->busFirstWordDelay = pkt->busLastWordDelay = 0;

        // queue the packet in the response queue to be sent out after
        // the static latency has passed
        port.schedTimingResp(pkt, curTick() + static_latency);
    } else {
        // @todo the packet is going to be deleted, and the DRAMPacket
        // is still having a pointer to it
        pendingDelete.push_back(pkt);
    }

    DPRINTF(DRAM, "Done\n");

    return;
}

void
DRAMCtrl::activateBank(Bank& bank, Tick act_tick, uint32_t row)
{
    // get the rank index from the bank
    uint8_t rank = bank.rank;

    assert(actTicks[rank].size() == activationLimit);

    DPRINTF(DRAM, "Activate at tick %d\n", act_tick);

    // update the open row
    assert(bank.openRow == Bank::NO_ROW);
    bank.openRow = row;

    // start counting anew, this covers both the case when we
    // auto-precharged, and when this access is forced to
    // precharge
    bank.bytesAccessed = 0;
    bank.rowAccesses = 0;

    ++numBanksActive;
    assert(numBanksActive <= banksPerRank * ranksPerChannel);

    DPRINTF(DRAM, "Activate bank %d, rank %d at tick %lld, now got %d active\n",
            bank.bank, bank.rank, act_tick, numBanksActive);

    DPRINTF(DRAMPower, "%llu,ACT,%d,%d\n", divCeil(act_tick, tCK), bank.bank,
            bank.rank);

    // The next access has to respect tRAS for this bank
    bank.preAllowedAt = act_tick + tRAS;

    // Respect the row-to-column command delay
    bank.colAllowedAt = act_tick + tRCD;

    // start by enforcing tRRD
    for(int i = 0; i < banksPerRank; i++) {
        // next activate to any bank in this rank must not happen
        // before tRRD
        banks[rank][i].actAllowedAt = std::max(act_tick + tRRD,
                                               banks[rank][i].actAllowedAt);
    }

    // next, we deal with tXAW, if the activation limit is disabled
    // then we are done
    if (actTicks[rank].empty())
        return;

    // sanity check
    if (actTicks[rank].back() && (act_tick - actTicks[rank].back()) < tXAW) {
        panic("Got %d activates in window %d (%llu - %llu) which is smaller "
              "than %llu\n", activationLimit, act_tick - actTicks[rank].back(),
              act_tick, actTicks[rank].back(), tXAW);
    }

    // shift the times used for the book keeping, the last element
    // (highest index) is the oldest one and hence the lowest value
    actTicks[rank].pop_back();

    // record an new activation (in the future)
    actTicks[rank].push_front(act_tick);

    // cannot activate more than X times in time window tXAW, push the
    // next one (the X + 1'st activate) to be tXAW away from the
    // oldest in our window of X
    if (actTicks[rank].back() && (act_tick - actTicks[rank].back()) < tXAW) {
        DPRINTF(DRAM, "Enforcing tXAW with X = %d, next activate no earlier "
                "than %llu\n", activationLimit, actTicks[rank].back() + tXAW);
            for(int j = 0; j < banksPerRank; j++)
                // next activate must not happen before end of window
                banks[rank][j].actAllowedAt =
                    std::max(actTicks[rank].back() + tXAW,
                             banks[rank][j].actAllowedAt);
    }

    // at the point when this activate takes place, make sure we
    // transition to the active power state
    if (!activateEvent.scheduled())
        schedule(activateEvent, act_tick);
    else if (activateEvent.when() > act_tick)
        // move it sooner in time
        reschedule(activateEvent, act_tick);
}

void
DRAMCtrl::processActivateEvent()
{
    // we should transition to the active state as soon as any bank is active
    if (pwrState != PWR_ACT)
        // note that at this point numBanksActive could be back at
        // zero again due to a precharge scheduled in the future
        schedulePowerEvent(PWR_ACT, curTick());
}

void
DRAMCtrl::prechargeBank(Bank& bank, Tick pre_at, bool trace)
{
    // make sure the bank has an open row
    assert(bank.openRow != Bank::NO_ROW);

    // sample the bytes per activate here since we are closing
    // the page
    bytesPerActivate.sample(bank.bytesAccessed);

    bank.openRow = Bank::NO_ROW;

    // no precharge allowed before this one
    bank.preAllowedAt = pre_at;

    Tick pre_done_at = pre_at + tRP;

    bank.actAllowedAt = std::max(bank.actAllowedAt, pre_done_at);

    assert(numBanksActive != 0);
    --numBanksActive;

    DPRINTF(DRAM, "Precharging bank %d, rank %d at tick %lld, now got "
            "%d active\n", bank.bank, bank.rank, pre_at, numBanksActive);

    if (trace)
        DPRINTF(DRAMPower, "%llu,PRE,%d,%d\n", divCeil(pre_at, tCK),
                bank.bank, bank.rank);

    // if we look at the current number of active banks we might be
    // tempted to think the DRAM is now idle, however this can be
    // undone by an activate that is scheduled to happen before we
    // would have reached the idle state, so schedule an event and
    // rather check once we actually make it to the point in time when
    // the (last) precharge takes place
    if (!prechargeEvent.scheduled())
        schedule(prechargeEvent, pre_done_at);
    else if (prechargeEvent.when() < pre_done_at)
        reschedule(prechargeEvent, pre_done_at);
}

void
DRAMCtrl::processPrechargeEvent()
{
    // if we reached zero, then special conditions apply as we track
    // if all banks are precharged for the power models
    if (numBanksActive == 0) {
        // we should transition to the idle state when the last bank
        // is precharged
        schedulePowerEvent(PWR_IDLE, curTick());
    }
}

void
DRAMCtrl::doDRAMAccess(DRAMPacket* dram_pkt)
{
    DPRINTF(DRAM, "Timing access to addr %lld, rank/bank/row %d %d %d\n",
            dram_pkt->addr, dram_pkt->rank, dram_pkt->bank, dram_pkt->row);

    // get the bank
    Bank& bank = dram_pkt->bankRef;

    // for the state we need to track if it is a row hit or not
    bool row_hit = true;

    // respect any constraints on the command (e.g. tRCD or tCCD)
    Tick cmd_at = std::max(bank.colAllowedAt, curTick());

    // Determine the access latency and update the bank state
    if (bank.openRow == dram_pkt->row) {
        // nothing to do
    } else {
        row_hit = false;

        // If there is a page open, precharge it.
        if (bank.openRow != Bank::NO_ROW) {
            prechargeBank(bank, std::max(bank.preAllowedAt, curTick()));
        }

        // next we need to account for the delay in activating the
        // page
        Tick act_tick = std::max(bank.actAllowedAt, curTick());

        // Record the activation and deal with all the global timing
        // constraints caused be a new activation (tRRD and tXAW)
        activateBank(bank, act_tick, dram_pkt->row);

        // issue the command as early as possible
        cmd_at = bank.colAllowedAt;
    }

    // we need to wait until the bus is available before we can issue
    // the command
    cmd_at = std::max(cmd_at, busBusyUntil - tCL);

    // update the packet ready time
    dram_pkt->readyTime = cmd_at + tCL + tBURST;

    // only one burst can use the bus at any one point in time
    assert(dram_pkt->readyTime - busBusyUntil >= tBURST);

    // not strictly necessary, but update the time for the next
    // read/write (add a max with tCCD here)
    bank.colAllowedAt = cmd_at + tBURST;

    // If this is a write, we also need to respect the write recovery
    // time before a precharge, in the case of a read, respect the
    // read to precharge constraint
    bank.preAllowedAt = std::max(bank.preAllowedAt,
                                 dram_pkt->isRead ? cmd_at + tRTP :
                                 dram_pkt->readyTime + tWR);

    // increment the bytes accessed and the accesses per row
    bank.bytesAccessed += burstSize;
    ++bank.rowAccesses;

    // if we reached the max, then issue with an auto-precharge
    bool auto_precharge = pageMgmt == Enums::close ||
        bank.rowAccesses == maxAccessesPerRow;

    // if we did not hit the limit, we might still want to
    // auto-precharge
    if (!auto_precharge &&
        (pageMgmt == Enums::open_adaptive ||
         pageMgmt == Enums::close_adaptive)) {
        // a twist on the open and close page policies:
        // 1) open_adaptive page policy does not blindly keep the
        // page open, but close it if there are no row hits, and there
        // are bank conflicts in the queue
        // 2) close_adaptive page policy does not blindly close the
        // page, but closes it only if there are no row hits in the queue.
        // In this case, only force an auto precharge when there
        // are no same page hits in the queue
        bool got_more_hits = false;
        bool got_bank_conflict = false;

        // either look at the read queue or write queue
        const deque<DRAMPacket*>& queue = dram_pkt->isRead ? readQueue :
            writeQueue;
        auto p = queue.begin();
        // make sure we are not considering the packet that we are
        // currently dealing with (which is the head of the queue)
        ++p;

        // keep on looking until we have found required condition or
        // reached the end
        while (!(got_more_hits &&
                 (got_bank_conflict || pageMgmt == Enums::close_adaptive)) &&
               p != queue.end()) {
            bool same_rank_bank = (dram_pkt->rank == (*p)->rank) &&
                (dram_pkt->bank == (*p)->bank);
            bool same_row = dram_pkt->row == (*p)->row;
            got_more_hits |= same_rank_bank && same_row;
            got_bank_conflict |= same_rank_bank && !same_row;
            ++p;
        }

        // auto pre-charge when either
        // 1) open_adaptive policy, we have not got any more hits, and
        //    have a bank conflict
        // 2) close_adaptive policy and we have not got any more hits
        auto_precharge = !got_more_hits &&
            (got_bank_conflict || pageMgmt == Enums::close_adaptive);
    }

    // DRAMPower trace command to be written
    std::string mem_cmd = dram_pkt->isRead ? "RD" : "WR";

    // if this access should use auto-precharge, then we are
    // closing the row
    if (auto_precharge) {
        prechargeBank(bank, std::max(curTick(), bank.preAllowedAt), false);

        mem_cmd.append("A");

        DPRINTF(DRAM, "Auto-precharged bank: %d\n", dram_pkt->bankId);
    }

    // Update bus state
    busBusyUntil = dram_pkt->readyTime;

    DPRINTF(DRAM, "Access to %lld, ready at %lld bus busy until %lld.\n",
            dram_pkt->addr, dram_pkt->readyTime, busBusyUntil);

    DPRINTF(DRAMPower, "%llu,%s,%d,%d\n", divCeil(cmd_at, tCK), mem_cmd,
            dram_pkt->bank, dram_pkt->rank);

    // Update the minimum timing between the requests, this is a
    // conservative estimate of when we have to schedule the next
    // request to not introduce any unecessary bubbles. In most cases
    // we will wake up sooner than we have to.
    nextReqTime = busBusyUntil - (tRP + tRCD + tCL);

    // Update the stats and schedule the next request
    if (dram_pkt->isRead) {
        ++readsThisTime;
        if (row_hit)
            readRowHits++;
        bytesReadDRAM += burstSize;
        perBankRdBursts[dram_pkt->bankId]++;

        // Update latency stats
        totMemAccLat += dram_pkt->readyTime - dram_pkt->entryTime;
        totBusLat += tBURST;
        totQLat += cmd_at - dram_pkt->entryTime;
    } else {
        ++writesThisTime;
        if (row_hit)
            writeRowHits++;
        bytesWritten += burstSize;
        perBankWrBursts[dram_pkt->bankId]++;
    }
}

void
DRAMCtrl::processNextReqEvent()
{
    if (busState == READ_TO_WRITE) {
        DPRINTF(DRAM, "Switching to writes after %d reads with %d reads "
                "waiting\n", readsThisTime, readQueue.size());

        // sample and reset the read-related stats as we are now
        // transitioning to writes, and all reads are done
        rdPerTurnAround.sample(readsThisTime);
        readsThisTime = 0;

        // now proceed to do the actual writes
        busState = WRITE;
    } else if (busState == WRITE_TO_READ) {
        DPRINTF(DRAM, "Switching to reads after %d writes with %d writes "
                "waiting\n", writesThisTime, writeQueue.size());

        wrPerTurnAround.sample(writesThisTime);
        writesThisTime = 0;

        busState = READ;
    }

    if (refreshState != REF_IDLE) {
        // if a refresh waiting for this event loop to finish, then hand
        // over now, and do not schedule a new nextReqEvent
        if (refreshState == REF_DRAIN) {
            DPRINTF(DRAM, "Refresh drain done, now precharging\n");

            refreshState = REF_PRE;

            // hand control back to the refresh event loop
            schedule(refreshEvent, curTick());
        }

        // let the refresh finish before issuing any further requests
        return;
    }

    // when we get here it is either a read or a write
    if (busState == READ) {

        // track if we should switch or not
        bool switch_to_writes = false;

        if (readQueue.empty()) {
            // In the case there is no read request to go next,
            // trigger writes if we have passed the low threshold (or
            // if we are draining)
            if (!writeQueue.empty() &&
                (drainManager || writeQueue.size() > writeLowThreshold)) {

                switch_to_writes = true;
            } else {
                // check if we are drained
                if (respQueue.empty () && drainManager) {
                    drainManager->signalDrainDone();
                    drainManager = NULL;
                }

                // nothing to do, not even any point in scheduling an
                // event for the next request
                return;
            }
        } else {
            // Figure out which read request goes next, and move it to the
            // front of the read queue
            chooseNext(readQueue);

            DRAMPacket* dram_pkt = readQueue.front();

            doDRAMAccess(dram_pkt);

            // At this point we're done dealing with the request
            readQueue.pop_front();

            // sanity check
            assert(dram_pkt->size <= burstSize);
            assert(dram_pkt->readyTime >= curTick());

            // Insert into response queue. It will be sent back to the
            // requestor at its readyTime
            if (respQueue.empty()) {
                assert(!respondEvent.scheduled());
                schedule(respondEvent, dram_pkt->readyTime);
            } else {
                assert(respQueue.back()->readyTime <= dram_pkt->readyTime);
                assert(respondEvent.scheduled());
            }

            respQueue.push_back(dram_pkt);

            // we have so many writes that we have to transition
            if (writeQueue.size() > writeHighThreshold) {
                switch_to_writes = true;
            }
        }

        // switching to writes, either because the read queue is empty
        // and the writes have passed the low threshold (or we are
        // draining), or because the writes hit the hight threshold
        if (switch_to_writes) {
            // transition to writing
            busState = READ_TO_WRITE;

            // add a bubble to the data bus, as defined by the
            // tRTW parameter
            busBusyUntil += tRTW;

            // update the minimum timing between the requests,
            // this shifts us back in time far enough to do any
            // bank preparation
            nextReqTime = busBusyUntil - (tRP + tRCD + tCL);
        }
    } else {
        chooseNext(writeQueue);
        DRAMPacket* dram_pkt = writeQueue.front();
        // sanity check
        assert(dram_pkt->size <= burstSize);
        doDRAMAccess(dram_pkt);

        writeQueue.pop_front();
        delete dram_pkt;

        // If we emptied the write queue, or got sufficiently below the
        // threshold (using the minWritesPerSwitch as the hysteresis) and
        // are not draining, or we have reads waiting and have done enough
        // writes, then switch to reads.
        if (writeQueue.empty() ||
            (writeQueue.size() + minWritesPerSwitch < writeLowThreshold &&
             !drainManager) ||
            (!readQueue.empty() && writesThisTime >= minWritesPerSwitch)) {
            // turn the bus back around for reads again
            busState = WRITE_TO_READ;

            // note that the we switch back to reads also in the idle
            // case, which eventually will check for any draining and
            // also pause any further scheduling if there is really
            // nothing to do

            // here we get a bit creative and shift the bus busy time not
            // just the tWTR, but also a CAS latency to capture the fact
            // that we are allowed to prepare a new bank, but not issue a
            // read command until after tWTR, in essence we capture a
            // bubble on the data bus that is tWTR + tCL
            busBusyUntil += tWTR + tCL;

            // update the minimum timing between the requests, this shifts
            // us back in time far enough to do any bank preparation
            nextReqTime = busBusyUntil - (tRP + tRCD + tCL);
        }
    }

    schedule(nextReqEvent, std::max(nextReqTime, curTick()));

    // If there is space available and we have writes waiting then let
    // them retry. This is done here to ensure that the retry does not
    // cause a nextReqEvent to be scheduled before we do so as part of
    // the next request processing
    if (retryWrReq && writeQueue.size() < writeBufferSize) {
        retryWrReq = false;
        port.sendRetry();
    }
}

uint64_t
DRAMCtrl::minBankActAt(const deque<DRAMPacket*>& queue) const
{
    uint64_t bank_mask = 0;
    Tick min_act_at = MaxTick;

    // deterimne if we have queued transactions targetting a
    // bank in question
    vector<bool> got_waiting(ranksPerChannel * banksPerRank, false);
    for (auto p = queue.begin(); p != queue.end(); ++p) {
        got_waiting[(*p)->bankId] = true;
    }

    for (int i = 0; i < ranksPerChannel; i++) {
        for (int j = 0; j < banksPerRank; j++) {
            uint8_t bank_id = i * banksPerRank + j;

            // if we have waiting requests for the bank, and it is
            // amongst the first available, update the mask
            if (got_waiting[bank_id]) {
                // simplistic approximation of when the bank can issue
                // an activate, ignoring any rank-to-rank switching
                // cost
                Tick act_at = banks[i][j].openRow == Bank::NO_ROW ?
                    banks[i][j].actAllowedAt :
                    std::max(banks[i][j].preAllowedAt, curTick()) + tRP;

                if (act_at <= min_act_at) {
                    // reset bank mask if new minimum is found
                    if (act_at < min_act_at)
                        bank_mask = 0;
                    // set the bit corresponding to the available bank
                    replaceBits(bank_mask, bank_id, bank_id, 1);
                    min_act_at = act_at;
                }
            }
        }
    }

    return bank_mask;
}

void
DRAMCtrl::processRefreshEvent()
{
    // when first preparing the refresh, remember when it was due
    if (refreshState == REF_IDLE) {
        // remember when the refresh is due
        refreshDueAt = curTick();

        // proceed to drain
        refreshState = REF_DRAIN;

        DPRINTF(DRAM, "Refresh due\n");
    }

    // let any scheduled read or write go ahead, after which it will
    // hand control back to this event loop
    if (refreshState == REF_DRAIN) {
        if (nextReqEvent.scheduled()) {
            // hand control over to the request loop until it is
            // evaluated next
            DPRINTF(DRAM, "Refresh awaiting draining\n");

            return;
        } else {
            refreshState = REF_PRE;
        }
    }

    // at this point, ensure that all banks are precharged
    if (refreshState == REF_PRE) {
        // precharge any active bank if we are not already in the idle
        // state
        if (pwrState != PWR_IDLE) {
            // at the moment, we use a precharge all even if there is
            // only a single bank open
            DPRINTF(DRAM, "Precharging all\n");

            // first determine when we can precharge
            Tick pre_at = curTick();
            for (int i = 0; i < ranksPerChannel; i++) {
                for (int j = 0; j < banksPerRank; j++) {
                    // respect both causality and any existing bank
                    // constraints, some banks could already have a
                    // (auto) precharge scheduled
                    pre_at = std::max(banks[i][j].preAllowedAt, pre_at);
                }
            }

            // make sure all banks are precharged, and for those that
            // already are, update their availability
            Tick act_allowed_at = pre_at + tRP;

            for (int i = 0; i < ranksPerChannel; i++) {
                for (int j = 0; j < banksPerRank; j++) {
                    if (banks[i][j].openRow != Bank::NO_ROW) {
                        prechargeBank(banks[i][j], pre_at, false);
                    } else {
                        banks[i][j].actAllowedAt =
                            std::max(banks[i][j].actAllowedAt, act_allowed_at);
                        banks[i][j].preAllowedAt =
                            std::max(banks[i][j].preAllowedAt, pre_at);
                    }
                }

                // at the moment this affects all ranks
                DPRINTF(DRAMPower, "%llu,PREA,0,%d\n", divCeil(pre_at, tCK),
                        i);
            }
        } else {
            DPRINTF(DRAM, "All banks already precharged, starting refresh\n");

            // go ahead and kick the power state machine into gear if
            // we are already idle
            schedulePowerEvent(PWR_REF, curTick());
        }

        refreshState = REF_RUN;
        assert(numBanksActive == 0);

        // wait for all banks to be precharged, at which point the
        // power state machine will transition to the idle state, and
        // automatically move to a refresh, at that point it will also
        // call this method to get the refresh event loop going again
        return;
    }

    // last but not least we perform the actual refresh
    if (refreshState == REF_RUN) {
        // should never get here with any banks active
        assert(numBanksActive == 0);
        assert(pwrState == PWR_REF);

        Tick ref_done_at = curTick() + tRFC;

        for (int i = 0; i < ranksPerChannel; i++) {
            for (int j = 0; j < banksPerRank; j++) {
                banks[i][j].actAllowedAt = ref_done_at;
            }

            // at the moment this affects all ranks
            DPRINTF(DRAMPower, "%llu,REF,0,%d\n", divCeil(curTick(), tCK), i);
        }

        // make sure we did not wait so long that we cannot make up
        // for it
        if (refreshDueAt + tREFI < ref_done_at) {
            fatal("Refresh was delayed so long we cannot catch up\n");
        }

        // compensate for the delay in actually performing the refresh
        // when scheduling the next one
        schedule(refreshEvent, refreshDueAt + tREFI - tRP);

        assert(!powerEvent.scheduled());

        // move to the idle power state once the refresh is done, this
        // will also move the refresh state machine to the refresh
        // idle state
        schedulePowerEvent(PWR_IDLE, ref_done_at);

        DPRINTF(DRAMState, "Refresh done at %llu and next refresh at %llu\n",
                ref_done_at, refreshDueAt + tREFI);
    }
}

void
DRAMCtrl::schedulePowerEvent(PowerState pwr_state, Tick tick)
{
    // respect causality
    assert(tick >= curTick());

    if (!powerEvent.scheduled()) {
        DPRINTF(DRAMState, "Scheduling power event at %llu to state %d\n",
                tick, pwr_state);

        // insert the new transition
        pwrStateTrans = pwr_state;

        schedule(powerEvent, tick);
    } else {
        panic("Scheduled power event at %llu to state %d, "
              "with scheduled event at %llu to %d\n", tick, pwr_state,
              powerEvent.when(), pwrStateTrans);
    }
}

void
DRAMCtrl::processPowerEvent()
{
    // remember where we were, and for how long
    Tick duration = curTick() - pwrStateTick;
    PowerState prev_state = pwrState;

    // update the accounting
    pwrStateTime[prev_state] += duration;

    pwrState = pwrStateTrans;
    pwrStateTick = curTick();

    if (pwrState == PWR_IDLE) {
        DPRINTF(DRAMState, "All banks precharged\n");

        // if we were refreshing, make sure we start scheduling requests again
        if (prev_state == PWR_REF) {
            DPRINTF(DRAMState, "Was refreshing for %llu ticks\n", duration);
            assert(pwrState == PWR_IDLE);

            // kick things into action again
            refreshState = REF_IDLE;
            assert(!nextReqEvent.scheduled());
            schedule(nextReqEvent, curTick());
        } else {
            assert(prev_state == PWR_ACT);

            // if we have a pending refresh, and are now moving to
            // the idle state, direclty transition to a refresh
            if (refreshState == REF_RUN) {
                // there should be nothing waiting at this point
                assert(!powerEvent.scheduled());

                // update the state in zero time and proceed below
                pwrState = PWR_REF;
            }
        }
    }

    // we transition to the refresh state, let the refresh state
    // machine know of this state update and let it deal with the
    // scheduling of the next power state transition as well as the
    // following refresh
    if (pwrState == PWR_REF) {
        DPRINTF(DRAMState, "Refreshing\n");
        // kick the refresh event loop into action again, and that
        // in turn will schedule a transition to the idle power
        // state once the refresh is done
        assert(refreshState == REF_RUN);
        processRefreshEvent();
    }
}

void
DRAMCtrl::regStats()
{
    using namespace Stats;

    AbstractMemory::regStats();

    readReqs
        .name(name() + ".readReqs")
        .desc("Number of read requests accepted");

    writeReqs
        .name(name() + ".writeReqs")
        .desc("Number of write requests accepted");

    readBursts
        .name(name() + ".readBursts")
        .desc("Number of DRAM read bursts, "
              "including those serviced by the write queue");

    writeBursts
        .name(name() + ".writeBursts")
        .desc("Number of DRAM write bursts, "
              "including those merged in the write queue");

    servicedByWrQ
        .name(name() + ".servicedByWrQ")
        .desc("Number of DRAM read bursts serviced by the write queue");

    mergedWrBursts
        .name(name() + ".mergedWrBursts")
        .desc("Number of DRAM write bursts merged with an existing one");

    neitherReadNorWrite
        .name(name() + ".neitherReadNorWriteReqs")
        .desc("Number of requests that are neither read nor write");

    perBankRdBursts
        .init(banksPerRank * ranksPerChannel)
        .name(name() + ".perBankRdBursts")
        .desc("Per bank write bursts");

    perBankWrBursts
        .init(banksPerRank * ranksPerChannel)
        .name(name() + ".perBankWrBursts")
        .desc("Per bank write bursts");

    avgRdQLen
        .name(name() + ".avgRdQLen")
        .desc("Average read queue length when enqueuing")
        .precision(2);

    avgWrQLen
        .name(name() + ".avgWrQLen")
        .desc("Average write queue length when enqueuing")
        .precision(2);

    totQLat
        .name(name() + ".totQLat")
        .desc("Total ticks spent queuing");

    totBusLat
        .name(name() + ".totBusLat")
        .desc("Total ticks spent in databus transfers");

    totMemAccLat
        .name(name() + ".totMemAccLat")
        .desc("Total ticks spent from burst creation until serviced "
              "by the DRAM");

    avgQLat
        .name(name() + ".avgQLat")
        .desc("Average queueing delay per DRAM burst")
        .precision(2);

    avgQLat = totQLat / (readBursts - servicedByWrQ);

    avgBusLat
        .name(name() + ".avgBusLat")
        .desc("Average bus latency per DRAM burst")
        .precision(2);

    avgBusLat = totBusLat / (readBursts - servicedByWrQ);

    avgMemAccLat
        .name(name() + ".avgMemAccLat")
        .desc("Average memory access latency per DRAM burst")
        .precision(2);

    avgMemAccLat = totMemAccLat / (readBursts - servicedByWrQ);

    numRdRetry
        .name(name() + ".numRdRetry")
        .desc("Number of times read queue was full causing retry");

    numWrRetry
        .name(name() + ".numWrRetry")
        .desc("Number of times write queue was full causing retry");

    readRowHits
        .name(name() + ".readRowHits")
        .desc("Number of row buffer hits during reads");

    writeRowHits
        .name(name() + ".writeRowHits")
        .desc("Number of row buffer hits during writes");

    readRowHitRate
        .name(name() + ".readRowHitRate")
        .desc("Row buffer hit rate for reads")
        .precision(2);

    readRowHitRate = (readRowHits / (readBursts - servicedByWrQ)) * 100;

    writeRowHitRate
        .name(name() + ".writeRowHitRate")
        .desc("Row buffer hit rate for writes")
        .precision(2);

    writeRowHitRate = (writeRowHits / (writeBursts - mergedWrBursts)) * 100;

    readPktSize
        .init(ceilLog2(burstSize) + 1)
        .name(name() + ".readPktSize")
        .desc("Read request sizes (log2)");

     writePktSize
        .init(ceilLog2(burstSize) + 1)
        .name(name() + ".writePktSize")
        .desc("Write request sizes (log2)");

     rdQLenPdf
        .init(readBufferSize)
        .name(name() + ".rdQLenPdf")
        .desc("What read queue length does an incoming req see");

     wrQLenPdf
        .init(writeBufferSize)
        .name(name() + ".wrQLenPdf")
        .desc("What write queue length does an incoming req see");

     bytesPerActivate
         .init(maxAccessesPerRow)
         .name(name() + ".bytesPerActivate")
         .desc("Bytes accessed per row activation")
         .flags(nozero);

     rdPerTurnAround
         .init(readBufferSize)
         .name(name() + ".rdPerTurnAround")
         .desc("Reads before turning the bus around for writes")
         .flags(nozero);

     wrPerTurnAround
         .init(writeBufferSize)
         .name(name() + ".wrPerTurnAround")
         .desc("Writes before turning the bus around for reads")
         .flags(nozero);

    bytesReadDRAM
        .name(name() + ".bytesReadDRAM")
        .desc("Total number of bytes read from DRAM");

    bytesReadWrQ
        .name(name() + ".bytesReadWrQ")
        .desc("Total number of bytes read from write queue");

    bytesWritten
        .name(name() + ".bytesWritten")
        .desc("Total number of bytes written to DRAM");

    bytesReadSys
        .name(name() + ".bytesReadSys")
        .desc("Total read bytes from the system interface side");

    bytesWrittenSys
        .name(name() + ".bytesWrittenSys")
        .desc("Total written bytes from the system interface side");

    avgRdBW
        .name(name() + ".avgRdBW")
        .desc("Average DRAM read bandwidth in MiByte/s")
        .precision(2);

    avgRdBW = (bytesReadDRAM / 1000000) / simSeconds;

    avgWrBW
        .name(name() + ".avgWrBW")
        .desc("Average achieved write bandwidth in MiByte/s")
        .precision(2);

    avgWrBW = (bytesWritten / 1000000) / simSeconds;

    avgRdBWSys
        .name(name() + ".avgRdBWSys")
        .desc("Average system read bandwidth in MiByte/s")
        .precision(2);

    avgRdBWSys = (bytesReadSys / 1000000) / simSeconds;

    avgWrBWSys
        .name(name() + ".avgWrBWSys")
        .desc("Average system write bandwidth in MiByte/s")
        .precision(2);

    avgWrBWSys = (bytesWrittenSys / 1000000) / simSeconds;

    peakBW
        .name(name() + ".peakBW")
        .desc("Theoretical peak bandwidth in MiByte/s")
        .precision(2);

    peakBW = (SimClock::Frequency / tBURST) * burstSize / 1000000;

    busUtil
        .name(name() + ".busUtil")
        .desc("Data bus utilization in percentage")
        .precision(2);

    busUtil = (avgRdBW + avgWrBW) / peakBW * 100;

    totGap
        .name(name() + ".totGap")
        .desc("Total gap between requests");

    avgGap
        .name(name() + ".avgGap")
        .desc("Average gap between requests")
        .precision(2);

    avgGap = totGap / (readReqs + writeReqs);

    // Stats for DRAM Power calculation based on Micron datasheet
    busUtilRead
        .name(name() + ".busUtilRead")
        .desc("Data bus utilization in percentage for reads")
        .precision(2);

    busUtilRead = avgRdBW / peakBW * 100;

    busUtilWrite
        .name(name() + ".busUtilWrite")
        .desc("Data bus utilization in percentage for writes")
        .precision(2);

    busUtilWrite = avgWrBW / peakBW * 100;

    pageHitRate
        .name(name() + ".pageHitRate")
        .desc("Row buffer hit rate, read and write combined")
        .precision(2);

    pageHitRate = (writeRowHits + readRowHits) /
        (writeBursts - mergedWrBursts + readBursts - servicedByWrQ) * 100;

    pwrStateTime
        .init(5)
        .name(name() + ".memoryStateTime")
        .desc("Time in different power states");
    pwrStateTime.subname(0, "IDLE");
    pwrStateTime.subname(1, "REF");
    pwrStateTime.subname(2, "PRE_PDN");
    pwrStateTime.subname(3, "ACT");
    pwrStateTime.subname(4, "ACT_PDN");
}

void
DRAMCtrl::recvFunctional(PacketPtr pkt)
{
    // rely on the abstract memory
    functionalAccess(pkt);
}

BaseSlavePort&
DRAMCtrl::getSlavePort(const string &if_name, PortID idx)
{
    if (if_name != "port") {
        return MemObject::getSlavePort(if_name, idx);
    } else {
        return port;
    }
}

unsigned int
DRAMCtrl::drain(DrainManager *dm)
{
    unsigned int count = port.drain(dm);

    // if there is anything in any of our internal queues, keep track
    // of that as well
    if (!(writeQueue.empty() && readQueue.empty() &&
          respQueue.empty())) {
        DPRINTF(Drain, "DRAM controller not drained, write: %d, read: %d,"
                " resp: %d\n", writeQueue.size(), readQueue.size(),
                respQueue.size());
        ++count;
        drainManager = dm;

        // the only part that is not drained automatically over time
        // is the write queue, thus kick things into action if needed
        if (!writeQueue.empty() && !nextReqEvent.scheduled()) {
            schedule(nextReqEvent, curTick());
        }
    }

    if (count)
        setDrainState(Drainable::Draining);
    else
        setDrainState(Drainable::Drained);
    return count;
}

DRAMCtrl::MemoryPort::MemoryPort(const std::string& name, DRAMCtrl& _memory)
    : QueuedSlavePort(name, &_memory, queue), queue(_memory, *this),
      memory(_memory)
{ }

AddrRangeList
DRAMCtrl::MemoryPort::getAddrRanges() const
{
    AddrRangeList ranges;
    ranges.push_back(memory.getAddrRange());
    return ranges;
}

void
DRAMCtrl::MemoryPort::recvFunctional(PacketPtr pkt)
{
    pkt->pushLabel(memory.name());

    if (!queue.checkFunctional(pkt)) {
        // Default implementation of SimpleTimingPort::recvFunctional()
        // calls recvAtomic() and throws away the latency; we can save a
        // little here by just not calculating the latency.
        memory.recvFunctional(pkt);
    }

    pkt->popLabel();
}

Tick
DRAMCtrl::MemoryPort::recvAtomic(PacketPtr pkt)
{
    return memory.recvAtomic(pkt);
}

bool
DRAMCtrl::MemoryPort::recvTimingReq(PacketPtr pkt)
{
    // pass it to the memory controller
    return memory.recvTimingReq(pkt);
}

DRAMCtrl*
DRAMCtrlParams::create()
{
    return new DRAMCtrl(this);
}