aboutsummaryrefslogtreecommitdiff
path: root/accel/tcg/cputlb.c
blob: defc8d59292037177c1180b4191448ee4df12dfe (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
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
/*
 *  Common CPU TLB handling
 *
 *  Copyright (c) 2003 Fabrice Bellard
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * This library is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
 */

#include "qemu/osdep.h"
#include "qemu/main-loop.h"
#include "cpu.h"
#include "exec/exec-all.h"
#include "exec/memory.h"
#include "exec/address-spaces.h"
#include "exec/cpu_ldst.h"
#include "exec/cputlb.h"
#include "exec/memory-internal.h"
#include "exec/ram_addr.h"
#include "tcg/tcg.h"
#include "qemu/error-report.h"
#include "exec/log.h"
#include "exec/helper-proto.h"
#include "qemu/atomic.h"
#include "qemu/atomic128.h"
#include "translate-all.h"

/* DEBUG defines, enable DEBUG_TLB_LOG to log to the CPU_LOG_MMU target */
/* #define DEBUG_TLB */
/* #define DEBUG_TLB_LOG */

#ifdef DEBUG_TLB
# define DEBUG_TLB_GATE 1
# ifdef DEBUG_TLB_LOG
#  define DEBUG_TLB_LOG_GATE 1
# else
#  define DEBUG_TLB_LOG_GATE 0
# endif
#else
# define DEBUG_TLB_GATE 0
# define DEBUG_TLB_LOG_GATE 0
#endif

#define tlb_debug(fmt, ...) do { \
    if (DEBUG_TLB_LOG_GATE) { \
        qemu_log_mask(CPU_LOG_MMU, "%s: " fmt, __func__, \
                      ## __VA_ARGS__); \
    } else if (DEBUG_TLB_GATE) { \
        fprintf(stderr, "%s: " fmt, __func__, ## __VA_ARGS__); \
    } \
} while (0)

#define assert_cpu_is_self(cpu) do {                              \
        if (DEBUG_TLB_GATE) {                                     \
            g_assert(!(cpu)->created || qemu_cpu_is_self(cpu));   \
        }                                                         \
    } while (0)

/* run_on_cpu_data.target_ptr should always be big enough for a
 * target_ulong even on 32 bit builds */
QEMU_BUILD_BUG_ON(sizeof(target_ulong) > sizeof(run_on_cpu_data));

/* We currently can't handle more than 16 bits in the MMUIDX bitmask.
 */
QEMU_BUILD_BUG_ON(NB_MMU_MODES > 16);
#define ALL_MMUIDX_BITS ((1 << NB_MMU_MODES) - 1)

static inline size_t sizeof_tlb(CPUArchState *env, uintptr_t mmu_idx)
{
    return env_tlb(env)->f[mmu_idx].mask + (1 << CPU_TLB_ENTRY_BITS);
}

static void tlb_window_reset(CPUTLBDesc *desc, int64_t ns,
                             size_t max_entries)
{
    desc->window_begin_ns = ns;
    desc->window_max_entries = max_entries;
}

static void tlb_dyn_init(CPUArchState *env)
{
    int i;

    for (i = 0; i < NB_MMU_MODES; i++) {
        CPUTLBDesc *desc = &env_tlb(env)->d[i];
        size_t n_entries = 1 << CPU_TLB_DYN_DEFAULT_BITS;

        tlb_window_reset(desc, get_clock_realtime(), 0);
        desc->n_used_entries = 0;
        env_tlb(env)->f[i].mask = (n_entries - 1) << CPU_TLB_ENTRY_BITS;
        env_tlb(env)->f[i].table = g_new(CPUTLBEntry, n_entries);
        env_tlb(env)->d[i].iotlb = g_new(CPUIOTLBEntry, n_entries);
    }
}

/**
 * tlb_mmu_resize_locked() - perform TLB resize bookkeeping; resize if necessary
 * @env: CPU that owns the TLB
 * @mmu_idx: MMU index of the TLB
 *
 * Called with tlb_lock_held.
 *
 * We have two main constraints when resizing a TLB: (1) we only resize it
 * on a TLB flush (otherwise we'd have to take a perf hit by either rehashing
 * the array or unnecessarily flushing it), which means we do not control how
 * frequently the resizing can occur; (2) we don't have access to the guest's
 * future scheduling decisions, and therefore have to decide the magnitude of
 * the resize based on past observations.
 *
 * In general, a memory-hungry process can benefit greatly from an appropriately
 * sized TLB, since a guest TLB miss is very expensive. This doesn't mean that
 * we just have to make the TLB as large as possible; while an oversized TLB
 * results in minimal TLB miss rates, it also takes longer to be flushed
 * (flushes can be _very_ frequent), and the reduced locality can also hurt
 * performance.
 *
 * To achieve near-optimal performance for all kinds of workloads, we:
 *
 * 1. Aggressively increase the size of the TLB when the use rate of the
 * TLB being flushed is high, since it is likely that in the near future this
 * memory-hungry process will execute again, and its memory hungriness will
 * probably be similar.
 *
 * 2. Slowly reduce the size of the TLB as the use rate declines over a
 * reasonably large time window. The rationale is that if in such a time window
 * we have not observed a high TLB use rate, it is likely that we won't observe
 * it in the near future. In that case, once a time window expires we downsize
 * the TLB to match the maximum use rate observed in the window.
 *
 * 3. Try to keep the maximum use rate in a time window in the 30-70% range,
 * since in that range performance is likely near-optimal. Recall that the TLB
 * is direct mapped, so we want the use rate to be low (or at least not too
 * high), since otherwise we are likely to have a significant amount of
 * conflict misses.
 */
static void tlb_mmu_resize_locked(CPUArchState *env, int mmu_idx)
{
    CPUTLBDesc *desc = &env_tlb(env)->d[mmu_idx];
    size_t old_size = tlb_n_entries(env, mmu_idx);
    size_t rate;
    size_t new_size = old_size;
    int64_t now = get_clock_realtime();
    int64_t window_len_ms = 100;
    int64_t window_len_ns = window_len_ms * 1000 * 1000;
    bool window_expired = now > desc->window_begin_ns + window_len_ns;

    if (desc->n_used_entries > desc->window_max_entries) {
        desc->window_max_entries = desc->n_used_entries;
    }
    rate = desc->window_max_entries * 100 / old_size;

    if (rate > 70) {
        new_size = MIN(old_size << 1, 1 << CPU_TLB_DYN_MAX_BITS);
    } else if (rate < 30 && window_expired) {
        size_t ceil = pow2ceil(desc->window_max_entries);
        size_t expected_rate = desc->window_max_entries * 100 / ceil;

        /*
         * Avoid undersizing when the max number of entries seen is just below
         * a pow2. For instance, if max_entries == 1025, the expected use rate
         * would be 1025/2048==50%. However, if max_entries == 1023, we'd get
         * 1023/1024==99.9% use rate, so we'd likely end up doubling the size
         * later. Thus, make sure that the expected use rate remains below 70%.
         * (and since we double the size, that means the lowest rate we'd
         * expect to get is 35%, which is still in the 30-70% range where
         * we consider that the size is appropriate.)
         */
        if (expected_rate > 70) {
            ceil *= 2;
        }
        new_size = MAX(ceil, 1 << CPU_TLB_DYN_MIN_BITS);
    }

    if (new_size == old_size) {
        if (window_expired) {
            tlb_window_reset(desc, now, desc->n_used_entries);
        }
        return;
    }

    g_free(env_tlb(env)->f[mmu_idx].table);
    g_free(env_tlb(env)->d[mmu_idx].iotlb);

    tlb_window_reset(desc, now, 0);
    /* desc->n_used_entries is cleared by the caller */
    env_tlb(env)->f[mmu_idx].mask = (new_size - 1) << CPU_TLB_ENTRY_BITS;
    env_tlb(env)->f[mmu_idx].table = g_try_new(CPUTLBEntry, new_size);
    env_tlb(env)->d[mmu_idx].iotlb = g_try_new(CPUIOTLBEntry, new_size);
    /*
     * If the allocations fail, try smaller sizes. We just freed some
     * memory, so going back to half of new_size has a good chance of working.
     * Increased memory pressure elsewhere in the system might cause the
     * allocations to fail though, so we progressively reduce the allocation
     * size, aborting if we cannot even allocate the smallest TLB we support.
     */
    while (env_tlb(env)->f[mmu_idx].table == NULL ||
           env_tlb(env)->d[mmu_idx].iotlb == NULL) {
        if (new_size == (1 << CPU_TLB_DYN_MIN_BITS)) {
            error_report("%s: %s", __func__, strerror(errno));
            abort();
        }
        new_size = MAX(new_size >> 1, 1 << CPU_TLB_DYN_MIN_BITS);
        env_tlb(env)->f[mmu_idx].mask = (new_size - 1) << CPU_TLB_ENTRY_BITS;

        g_free(env_tlb(env)->f[mmu_idx].table);
        g_free(env_tlb(env)->d[mmu_idx].iotlb);
        env_tlb(env)->f[mmu_idx].table = g_try_new(CPUTLBEntry, new_size);
        env_tlb(env)->d[mmu_idx].iotlb = g_try_new(CPUIOTLBEntry, new_size);
    }
}

static inline void tlb_table_flush_by_mmuidx(CPUArchState *env, int mmu_idx)
{
    tlb_mmu_resize_locked(env, mmu_idx);
    memset(env_tlb(env)->f[mmu_idx].table, -1, sizeof_tlb(env, mmu_idx));
    env_tlb(env)->d[mmu_idx].n_used_entries = 0;
}

static inline void tlb_n_used_entries_inc(CPUArchState *env, uintptr_t mmu_idx)
{
    env_tlb(env)->d[mmu_idx].n_used_entries++;
}

static inline void tlb_n_used_entries_dec(CPUArchState *env, uintptr_t mmu_idx)
{
    env_tlb(env)->d[mmu_idx].n_used_entries--;
}

void tlb_init(CPUState *cpu)
{
    CPUArchState *env = cpu->env_ptr;

    qemu_spin_init(&env_tlb(env)->c.lock);

    /* Ensure that cpu_reset performs a full flush.  */
    env_tlb(env)->c.dirty = ALL_MMUIDX_BITS;

    tlb_dyn_init(env);
}

/* flush_all_helper: run fn across all cpus
 *
 * If the wait flag is set then the src cpu's helper will be queued as
 * "safe" work and the loop exited creating a synchronisation point
 * where all queued work will be finished before execution starts
 * again.
 */
static void flush_all_helper(CPUState *src, run_on_cpu_func fn,
                             run_on_cpu_data d)
{
    CPUState *cpu;

    CPU_FOREACH(cpu) {
        if (cpu != src) {
            async_run_on_cpu(cpu, fn, d);
        }
    }
}

void tlb_flush_counts(size_t *pfull, size_t *ppart, size_t *pelide)
{
    CPUState *cpu;
    size_t full = 0, part = 0, elide = 0;

    CPU_FOREACH(cpu) {
        CPUArchState *env = cpu->env_ptr;

        full += atomic_read(&env_tlb(env)->c.full_flush_count);
        part += atomic_read(&env_tlb(env)->c.part_flush_count);
        elide += atomic_read(&env_tlb(env)->c.elide_flush_count);
    }
    *pfull = full;
    *ppart = part;
    *pelide = elide;
}

static void tlb_flush_one_mmuidx_locked(CPUArchState *env, int mmu_idx)
{
    tlb_table_flush_by_mmuidx(env, mmu_idx);
    env_tlb(env)->d[mmu_idx].large_page_addr = -1;
    env_tlb(env)->d[mmu_idx].large_page_mask = -1;
    env_tlb(env)->d[mmu_idx].vindex = 0;
    memset(env_tlb(env)->d[mmu_idx].vtable, -1,
           sizeof(env_tlb(env)->d[0].vtable));
}

static void tlb_flush_by_mmuidx_async_work(CPUState *cpu, run_on_cpu_data data)
{
    CPUArchState *env = cpu->env_ptr;
    uint16_t asked = data.host_int;
    uint16_t all_dirty, work, to_clean;

    assert_cpu_is_self(cpu);

    tlb_debug("mmu_idx:0x%04" PRIx16 "\n", asked);

    qemu_spin_lock(&env_tlb(env)->c.lock);

    all_dirty = env_tlb(env)->c.dirty;
    to_clean = asked & all_dirty;
    all_dirty &= ~to_clean;
    env_tlb(env)->c.dirty = all_dirty;

    for (work = to_clean; work != 0; work &= work - 1) {
        int mmu_idx = ctz32(work);
        tlb_flush_one_mmuidx_locked(env, mmu_idx);
    }

    qemu_spin_unlock(&env_tlb(env)->c.lock);

    cpu_tb_jmp_cache_clear(cpu);

    if (to_clean == ALL_MMUIDX_BITS) {
        atomic_set(&env_tlb(env)->c.full_flush_count,
                   env_tlb(env)->c.full_flush_count + 1);
    } else {
        atomic_set(&env_tlb(env)->c.part_flush_count,
                   env_tlb(env)->c.part_flush_count + ctpop16(to_clean));
        if (to_clean != asked) {
            atomic_set(&env_tlb(env)->c.elide_flush_count,
                       env_tlb(env)->c.elide_flush_count +
                       ctpop16(asked & ~to_clean));
        }
    }
}

void tlb_flush_by_mmuidx(CPUState *cpu, uint16_t idxmap)
{
    tlb_debug("mmu_idx: 0x%" PRIx16 "\n", idxmap);

    if (cpu->created && !qemu_cpu_is_self(cpu)) {
        async_run_on_cpu(cpu, tlb_flush_by_mmuidx_async_work,
                         RUN_ON_CPU_HOST_INT(idxmap));
    } else {
        tlb_flush_by_mmuidx_async_work(cpu, RUN_ON_CPU_HOST_INT(idxmap));
    }
}

void tlb_flush(CPUState *cpu)
{
    tlb_flush_by_mmuidx(cpu, ALL_MMUIDX_BITS);
}

void tlb_flush_by_mmuidx_all_cpus(CPUState *src_cpu, uint16_t idxmap)
{
    const run_on_cpu_func fn = tlb_flush_by_mmuidx_async_work;

    tlb_debug("mmu_idx: 0x%"PRIx16"\n", idxmap);

    flush_all_helper(src_cpu, fn, RUN_ON_CPU_HOST_INT(idxmap));
    fn(src_cpu, RUN_ON_CPU_HOST_INT(idxmap));
}

void tlb_flush_all_cpus(CPUState *src_cpu)
{
    tlb_flush_by_mmuidx_all_cpus(src_cpu, ALL_MMUIDX_BITS);
}

void tlb_flush_by_mmuidx_all_cpus_synced(CPUState *src_cpu, uint16_t idxmap)
{
    const run_on_cpu_func fn = tlb_flush_by_mmuidx_async_work;

    tlb_debug("mmu_idx: 0x%"PRIx16"\n", idxmap);

    flush_all_helper(src_cpu, fn, RUN_ON_CPU_HOST_INT(idxmap));
    async_safe_run_on_cpu(src_cpu, fn, RUN_ON_CPU_HOST_INT(idxmap));
}

void tlb_flush_all_cpus_synced(CPUState *src_cpu)
{
    tlb_flush_by_mmuidx_all_cpus_synced(src_cpu, ALL_MMUIDX_BITS);
}

static inline bool tlb_hit_page_anyprot(CPUTLBEntry *tlb_entry,
                                        target_ulong page)
{
    return tlb_hit_page(tlb_entry->addr_read, page) ||
           tlb_hit_page(tlb_addr_write(tlb_entry), page) ||
           tlb_hit_page(tlb_entry->addr_code, page);
}

/**
 * tlb_entry_is_empty - return true if the entry is not in use
 * @te: pointer to CPUTLBEntry
 */
static inline bool tlb_entry_is_empty(const CPUTLBEntry *te)
{
    return te->addr_read == -1 && te->addr_write == -1 && te->addr_code == -1;
}

/* Called with tlb_c.lock held */
static inline bool tlb_flush_entry_locked(CPUTLBEntry *tlb_entry,
                                          target_ulong page)
{
    if (tlb_hit_page_anyprot(tlb_entry, page)) {
        memset(tlb_entry, -1, sizeof(*tlb_entry));
        return true;
    }
    return false;
}

/* Called with tlb_c.lock held */
static inline void tlb_flush_vtlb_page_locked(CPUArchState *env, int mmu_idx,
                                              target_ulong page)
{
    CPUTLBDesc *d = &env_tlb(env)->d[mmu_idx];
    int k;

    assert_cpu_is_self(env_cpu(env));
    for (k = 0; k < CPU_VTLB_SIZE; k++) {
        if (tlb_flush_entry_locked(&d->vtable[k], page)) {
            tlb_n_used_entries_dec(env, mmu_idx);
        }
    }
}

static void tlb_flush_page_locked(CPUArchState *env, int midx,
                                  target_ulong page)
{
    target_ulong lp_addr = env_tlb(env)->d[midx].large_page_addr;
    target_ulong lp_mask = env_tlb(env)->d[midx].large_page_mask;

    /* Check if we need to flush due to large pages.  */
    if ((page & lp_mask) == lp_addr) {
        tlb_debug("forcing full flush midx %d ("
                  TARGET_FMT_lx "/" TARGET_FMT_lx ")\n",
                  midx, lp_addr, lp_mask);
        tlb_flush_one_mmuidx_locked(env, midx);
    } else {
        if (tlb_flush_entry_locked(tlb_entry(env, midx, page), page)) {
            tlb_n_used_entries_dec(env, midx);
        }
        tlb_flush_vtlb_page_locked(env, midx, page);
    }
}

/* As we are going to hijack the bottom bits of the page address for a
 * mmuidx bit mask we need to fail to build if we can't do that
 */
QEMU_BUILD_BUG_ON(NB_MMU_MODES > TARGET_PAGE_BITS_MIN);

static void tlb_flush_page_by_mmuidx_async_work(CPUState *cpu,
                                                run_on_cpu_data data)
{
    CPUArchState *env = cpu->env_ptr;
    target_ulong addr_and_mmuidx = (target_ulong) data.target_ptr;
    target_ulong addr = addr_and_mmuidx & TARGET_PAGE_MASK;
    unsigned long mmu_idx_bitmap = addr_and_mmuidx & ALL_MMUIDX_BITS;
    int mmu_idx;

    assert_cpu_is_self(cpu);

    tlb_debug("page addr:" TARGET_FMT_lx " mmu_map:0x%lx\n",
              addr, mmu_idx_bitmap);

    qemu_spin_lock(&env_tlb(env)->c.lock);
    for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
        if (test_bit(mmu_idx, &mmu_idx_bitmap)) {
            tlb_flush_page_locked(env, mmu_idx, addr);
        }
    }
    qemu_spin_unlock(&env_tlb(env)->c.lock);

    tb_flush_jmp_cache(cpu, addr);
}

void tlb_flush_page_by_mmuidx(CPUState *cpu, target_ulong addr, uint16_t idxmap)
{
    target_ulong addr_and_mmu_idx;

    tlb_debug("addr: "TARGET_FMT_lx" mmu_idx:%" PRIx16 "\n", addr, idxmap);

    /* This should already be page aligned */
    addr_and_mmu_idx = addr & TARGET_PAGE_MASK;
    addr_and_mmu_idx |= idxmap;

    if (!qemu_cpu_is_self(cpu)) {
        async_run_on_cpu(cpu, tlb_flush_page_by_mmuidx_async_work,
                         RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
    } else {
        tlb_flush_page_by_mmuidx_async_work(
            cpu, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
    }
}

void tlb_flush_page(CPUState *cpu, target_ulong addr)
{
    tlb_flush_page_by_mmuidx(cpu, addr, ALL_MMUIDX_BITS);
}

void tlb_flush_page_by_mmuidx_all_cpus(CPUState *src_cpu, target_ulong addr,
                                       uint16_t idxmap)
{
    const run_on_cpu_func fn = tlb_flush_page_by_mmuidx_async_work;
    target_ulong addr_and_mmu_idx;

    tlb_debug("addr: "TARGET_FMT_lx" mmu_idx:%"PRIx16"\n", addr, idxmap);

    /* This should already be page aligned */
    addr_and_mmu_idx = addr & TARGET_PAGE_MASK;
    addr_and_mmu_idx |= idxmap;

    flush_all_helper(src_cpu, fn, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
    fn(src_cpu, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
}

void tlb_flush_page_all_cpus(CPUState *src, target_ulong addr)
{
    tlb_flush_page_by_mmuidx_all_cpus(src, addr, ALL_MMUIDX_BITS);
}

void tlb_flush_page_by_mmuidx_all_cpus_synced(CPUState *src_cpu,
                                              target_ulong addr,
                                              uint16_t idxmap)
{
    const run_on_cpu_func fn = tlb_flush_page_by_mmuidx_async_work;
    target_ulong addr_and_mmu_idx;

    tlb_debug("addr: "TARGET_FMT_lx" mmu_idx:%"PRIx16"\n", addr, idxmap);

    /* This should already be page aligned */
    addr_and_mmu_idx = addr & TARGET_PAGE_MASK;
    addr_and_mmu_idx |= idxmap;

    flush_all_helper(src_cpu, fn, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
    async_safe_run_on_cpu(src_cpu, fn, RUN_ON_CPU_TARGET_PTR(addr_and_mmu_idx));
}

void tlb_flush_page_all_cpus_synced(CPUState *src, target_ulong addr)
{
    tlb_flush_page_by_mmuidx_all_cpus_synced(src, addr, ALL_MMUIDX_BITS);
}

/* update the TLBs so that writes to code in the virtual page 'addr'
   can be detected */
void tlb_protect_code(ram_addr_t ram_addr)
{
    cpu_physical_memory_test_and_clear_dirty(ram_addr, TARGET_PAGE_SIZE,
                                             DIRTY_MEMORY_CODE);
}

/* update the TLB so that writes in physical page 'phys_addr' are no longer
   tested for self modifying code */
void tlb_unprotect_code(ram_addr_t ram_addr)
{
    cpu_physical_memory_set_dirty_flag(ram_addr, DIRTY_MEMORY_CODE);
}


/*
 * Dirty write flag handling
 *
 * When the TCG code writes to a location it looks up the address in
 * the TLB and uses that data to compute the final address. If any of
 * the lower bits of the address are set then the slow path is forced.
 * There are a number of reasons to do this but for normal RAM the
 * most usual is detecting writes to code regions which may invalidate
 * generated code.
 *
 * Other vCPUs might be reading their TLBs during guest execution, so we update
 * te->addr_write with atomic_set. We don't need to worry about this for
 * oversized guests as MTTCG is disabled for them.
 *
 * Called with tlb_c.lock held.
 */
static void tlb_reset_dirty_range_locked(CPUTLBEntry *tlb_entry,
                                         uintptr_t start, uintptr_t length)
{
    uintptr_t addr = tlb_entry->addr_write;

    if ((addr & (TLB_INVALID_MASK | TLB_MMIO |
                 TLB_DISCARD_WRITE | TLB_NOTDIRTY)) == 0) {
        addr &= TARGET_PAGE_MASK;
        addr += tlb_entry->addend;
        if ((addr - start) < length) {
#if TCG_OVERSIZED_GUEST
            tlb_entry->addr_write |= TLB_NOTDIRTY;
#else
            atomic_set(&tlb_entry->addr_write,
                       tlb_entry->addr_write | TLB_NOTDIRTY);
#endif
        }
    }
}

/*
 * Called with tlb_c.lock held.
 * Called only from the vCPU context, i.e. the TLB's owner thread.
 */
static inline void copy_tlb_helper_locked(CPUTLBEntry *d, const CPUTLBEntry *s)
{
    *d = *s;
}

/* This is a cross vCPU call (i.e. another vCPU resetting the flags of
 * the target vCPU).
 * We must take tlb_c.lock to avoid racing with another vCPU update. The only
 * thing actually updated is the target TLB entry ->addr_write flags.
 */
void tlb_reset_dirty(CPUState *cpu, ram_addr_t start1, ram_addr_t length)
{
    CPUArchState *env;

    int mmu_idx;

    env = cpu->env_ptr;
    qemu_spin_lock(&env_tlb(env)->c.lock);
    for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
        unsigned int i;
        unsigned int n = tlb_n_entries(env, mmu_idx);

        for (i = 0; i < n; i++) {
            tlb_reset_dirty_range_locked(&env_tlb(env)->f[mmu_idx].table[i],
                                         start1, length);
        }

        for (i = 0; i < CPU_VTLB_SIZE; i++) {
            tlb_reset_dirty_range_locked(&env_tlb(env)->d[mmu_idx].vtable[i],
                                         start1, length);
        }
    }
    qemu_spin_unlock(&env_tlb(env)->c.lock);
}

/* Called with tlb_c.lock held */
static inline void tlb_set_dirty1_locked(CPUTLBEntry *tlb_entry,
                                         target_ulong vaddr)
{
    if (tlb_entry->addr_write == (vaddr | TLB_NOTDIRTY)) {
        tlb_entry->addr_write = vaddr;
    }
}

/* update the TLB corresponding to virtual page vaddr
   so that it is no longer dirty */
void tlb_set_dirty(CPUState *cpu, target_ulong vaddr)
{
    CPUArchState *env = cpu->env_ptr;
    int mmu_idx;

    assert_cpu_is_self(cpu);

    vaddr &= TARGET_PAGE_MASK;
    qemu_spin_lock(&env_tlb(env)->c.lock);
    for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
        tlb_set_dirty1_locked(tlb_entry(env, mmu_idx, vaddr), vaddr);
    }

    for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
        int k;
        for (k = 0; k < CPU_VTLB_SIZE; k++) {
            tlb_set_dirty1_locked(&env_tlb(env)->d[mmu_idx].vtable[k], vaddr);
        }
    }
    qemu_spin_unlock(&env_tlb(env)->c.lock);
}

/* Our TLB does not support large pages, so remember the area covered by
   large pages and trigger a full TLB flush if these are invalidated.  */
static void tlb_add_large_page(CPUArchState *env, int mmu_idx,
                               target_ulong vaddr, target_ulong size)
{
    target_ulong lp_addr = env_tlb(env)->d[mmu_idx].large_page_addr;
    target_ulong lp_mask = ~(size - 1);

    if (lp_addr == (target_ulong)-1) {
        /* No previous large page.  */
        lp_addr = vaddr;
    } else {
        /* Extend the existing region to include the new page.
           This is a compromise between unnecessary flushes and
           the cost of maintaining a full variable size TLB.  */
        lp_mask &= env_tlb(env)->d[mmu_idx].large_page_mask;
        while (((lp_addr ^ vaddr) & lp_mask) != 0) {
            lp_mask <<= 1;
        }
    }
    env_tlb(env)->d[mmu_idx].large_page_addr = lp_addr & lp_mask;
    env_tlb(env)->d[mmu_idx].large_page_mask = lp_mask;
}

/* Add a new TLB entry. At most one entry for a given virtual address
 * is permitted. Only a single TARGET_PAGE_SIZE region is mapped, the
 * supplied size is only used by tlb_flush_page.
 *
 * Called from TCG-generated code, which is under an RCU read-side
 * critical section.
 */
void tlb_set_page_with_attrs(CPUState *cpu, target_ulong vaddr,
                             hwaddr paddr, MemTxAttrs attrs, int prot,
                             int mmu_idx, target_ulong size)
{
    CPUArchState *env = cpu->env_ptr;
    CPUTLB *tlb = env_tlb(env);
    CPUTLBDesc *desc = &tlb->d[mmu_idx];
    MemoryRegionSection *section;
    unsigned int index;
    target_ulong address;
    target_ulong write_address;
    uintptr_t addend;
    CPUTLBEntry *te, tn;
    hwaddr iotlb, xlat, sz, paddr_page;
    target_ulong vaddr_page;
    int asidx = cpu_asidx_from_attrs(cpu, attrs);
    int wp_flags;
    bool is_ram, is_romd;

    assert_cpu_is_self(cpu);

    if (size <= TARGET_PAGE_SIZE) {
        sz = TARGET_PAGE_SIZE;
    } else {
        tlb_add_large_page(env, mmu_idx, vaddr, size);
        sz = size;
    }
    vaddr_page = vaddr & TARGET_PAGE_MASK;
    paddr_page = paddr & TARGET_PAGE_MASK;

    section = address_space_translate_for_iotlb(cpu, asidx, paddr_page,
                                                &xlat, &sz, attrs, &prot);
    assert(sz >= TARGET_PAGE_SIZE);

    tlb_debug("vaddr=" TARGET_FMT_lx " paddr=0x" TARGET_FMT_plx
              " prot=%x idx=%d\n",
              vaddr, paddr, prot, mmu_idx);

    address = vaddr_page;
    if (size < TARGET_PAGE_SIZE) {
        /* Repeat the MMU check and TLB fill on every access.  */
        address |= TLB_INVALID_MASK;
    }
    if (attrs.byte_swap) {
        address |= TLB_BSWAP;
    }

    is_ram = memory_region_is_ram(section->mr);
    is_romd = memory_region_is_romd(section->mr);

    if (is_ram || is_romd) {
        /* RAM and ROMD both have associated host memory. */
        addend = (uintptr_t)memory_region_get_ram_ptr(section->mr) + xlat;
    } else {
        /* I/O does not; force the host address to NULL. */
        addend = 0;
    }

    write_address = address;
    if (is_ram) {
        iotlb = memory_region_get_ram_addr(section->mr) + xlat;
        /*
         * Computing is_clean is expensive; avoid all that unless
         * the page is actually writable.
         */
        if (prot & PAGE_WRITE) {
            if (section->readonly) {
                write_address |= TLB_DISCARD_WRITE;
            } else if (cpu_physical_memory_is_clean(iotlb)) {
                write_address |= TLB_NOTDIRTY;
            }
        }
    } else {
        /* I/O or ROMD */
        iotlb = memory_region_section_get_iotlb(cpu, section) + xlat;
        /*
         * Writes to romd devices must go through MMIO to enable write.
         * Reads to romd devices go through the ram_ptr found above,
         * but of course reads to I/O must go through MMIO.
         */
        write_address |= TLB_MMIO;
        if (!is_romd) {
            address = write_address;
        }
    }

    wp_flags = cpu_watchpoint_address_matches(cpu, vaddr_page,
                                              TARGET_PAGE_SIZE);

    index = tlb_index(env, mmu_idx, vaddr_page);
    te = tlb_entry(env, mmu_idx, vaddr_page);

    /*
     * Hold the TLB lock for the rest of the function. We could acquire/release
     * the lock several times in the function, but it is faster to amortize the
     * acquisition cost by acquiring it just once. Note that this leads to
     * a longer critical section, but this is not a concern since the TLB lock
     * is unlikely to be contended.
     */
    qemu_spin_lock(&tlb->c.lock);

    /* Note that the tlb is no longer clean.  */
    tlb->c.dirty |= 1 << mmu_idx;

    /* Make sure there's no cached translation for the new page.  */
    tlb_flush_vtlb_page_locked(env, mmu_idx, vaddr_page);

    /*
     * Only evict the old entry to the victim tlb if it's for a
     * different page; otherwise just overwrite the stale data.
     */
    if (!tlb_hit_page_anyprot(te, vaddr_page) && !tlb_entry_is_empty(te)) {
        unsigned vidx = desc->vindex++ % CPU_VTLB_SIZE;
        CPUTLBEntry *tv = &desc->vtable[vidx];

        /* Evict the old entry into the victim tlb.  */
        copy_tlb_helper_locked(tv, te);
        desc->viotlb[vidx] = desc->iotlb[index];
        tlb_n_used_entries_dec(env, mmu_idx);
    }

    /* refill the tlb */
    /*
     * At this point iotlb contains a physical section number in the lower
     * TARGET_PAGE_BITS, and either
     *  + the ram_addr_t of the page base of the target RAM (RAM)
     *  + the offset within section->mr of the page base (I/O, ROMD)
     * We subtract the vaddr_page (which is page aligned and thus won't
     * disturb the low bits) to give an offset which can be added to the
     * (non-page-aligned) vaddr of the eventual memory access to get
     * the MemoryRegion offset for the access. Note that the vaddr we
     * subtract here is that of the page base, and not the same as the
     * vaddr we add back in io_readx()/io_writex()/get_page_addr_code().
     */
    desc->iotlb[index].addr = iotlb - vaddr_page;
    desc->iotlb[index].attrs = attrs;

    /* Now calculate the new entry */
    tn.addend = addend - vaddr_page;
    if (prot & PAGE_READ) {
        tn.addr_read = address;
        if (wp_flags & BP_MEM_READ) {
            tn.addr_read |= TLB_WATCHPOINT;
        }
    } else {
        tn.addr_read = -1;
    }

    if (prot & PAGE_EXEC) {
        tn.addr_code = address;
    } else {
        tn.addr_code = -1;
    }

    tn.addr_write = -1;
    if (prot & PAGE_WRITE) {
        tn.addr_write = write_address;
        if (prot & PAGE_WRITE_INV) {
            tn.addr_write |= TLB_INVALID_MASK;
        }
        if (wp_flags & BP_MEM_WRITE) {
            tn.addr_write |= TLB_WATCHPOINT;
        }
    }

    copy_tlb_helper_locked(te, &tn);
    tlb_n_used_entries_inc(env, mmu_idx);
    qemu_spin_unlock(&tlb->c.lock);
}

/* Add a new TLB entry, but without specifying the memory
 * transaction attributes to be used.
 */
void tlb_set_page(CPUState *cpu, target_ulong vaddr,
                  hwaddr paddr, int prot,
                  int mmu_idx, target_ulong size)
{
    tlb_set_page_with_attrs(cpu, vaddr, paddr, MEMTXATTRS_UNSPECIFIED,
                            prot, mmu_idx, size);
}

static inline ram_addr_t qemu_ram_addr_from_host_nofail(void *ptr)
{
    ram_addr_t ram_addr;

    ram_addr = qemu_ram_addr_from_host(ptr);
    if (ram_addr == RAM_ADDR_INVALID) {
        error_report("Bad ram pointer %p", ptr);
        abort();
    }
    return ram_addr;
}

/*
 * Note: tlb_fill() can trigger a resize of the TLB. This means that all of the
 * caller's prior references to the TLB table (e.g. CPUTLBEntry pointers) must
 * be discarded and looked up again (e.g. via tlb_entry()).
 */
static void tlb_fill(CPUState *cpu, target_ulong addr, int size,
                     MMUAccessType access_type, int mmu_idx, uintptr_t retaddr)
{
    CPUClass *cc = CPU_GET_CLASS(cpu);
    bool ok;

    /*
     * This is not a probe, so only valid return is success; failure
     * should result in exception + longjmp to the cpu loop.
     */
    ok = cc->tlb_fill(cpu, addr, size, access_type, mmu_idx, false, retaddr);
    assert(ok);
}

static uint64_t io_readx(CPUArchState *env, CPUIOTLBEntry *iotlbentry,
                         int mmu_idx, target_ulong addr, uintptr_t retaddr,
                         MMUAccessType access_type, MemOp op)
{
    CPUState *cpu = env_cpu(env);
    hwaddr mr_offset;
    MemoryRegionSection *section;
    MemoryRegion *mr;
    uint64_t val;
    bool locked = false;
    MemTxResult r;

    section = iotlb_to_section(cpu, iotlbentry->addr, iotlbentry->attrs);
    mr = section->mr;
    mr_offset = (iotlbentry->addr & TARGET_PAGE_MASK) + addr;
    cpu->mem_io_pc = retaddr;
    if (!cpu->can_do_io) {
        cpu_io_recompile(cpu, retaddr);
    }

    cpu->mem_io_access_type = access_type;

    if (mr->global_locking && !qemu_mutex_iothread_locked()) {
        qemu_mutex_lock_iothread();
        locked = true;
    }
    r = memory_region_dispatch_read(mr, mr_offset, &val, op, iotlbentry->attrs);
    if (r != MEMTX_OK) {
        hwaddr physaddr = mr_offset +
            section->offset_within_address_space -
            section->offset_within_region;

        cpu_transaction_failed(cpu, physaddr, addr, memop_size(op), access_type,
                               mmu_idx, iotlbentry->attrs, r, retaddr);
    }
    if (locked) {
        qemu_mutex_unlock_iothread();
    }

    return val;
}

static void io_writex(CPUArchState *env, CPUIOTLBEntry *iotlbentry,
                      int mmu_idx, uint64_t val, target_ulong addr,
                      uintptr_t retaddr, MemOp op)
{
    CPUState *cpu = env_cpu(env);
    hwaddr mr_offset;
    MemoryRegionSection *section;
    MemoryRegion *mr;
    bool locked = false;
    MemTxResult r;

    section = iotlb_to_section(cpu, iotlbentry->addr, iotlbentry->attrs);
    mr = section->mr;
    mr_offset = (iotlbentry->addr & TARGET_PAGE_MASK) + addr;
    if (!cpu->can_do_io) {
        cpu_io_recompile(cpu, retaddr);
    }
    cpu->mem_io_pc = retaddr;

    if (mr->global_locking && !qemu_mutex_iothread_locked()) {
        qemu_mutex_lock_iothread();
        locked = true;
    }
    r = memory_region_dispatch_write(mr, mr_offset, val, op, iotlbentry->attrs);
    if (r != MEMTX_OK) {
        hwaddr physaddr = mr_offset +
            section->offset_within_address_space -
            section->offset_within_region;

        cpu_transaction_failed(cpu, physaddr, addr, memop_size(op),
                               MMU_DATA_STORE, mmu_idx, iotlbentry->attrs, r,
                               retaddr);
    }
    if (locked) {
        qemu_mutex_unlock_iothread();
    }
}

static inline target_ulong tlb_read_ofs(CPUTLBEntry *entry, size_t ofs)
{
#if TCG_OVERSIZED_GUEST
    return *(target_ulong *)((uintptr_t)entry + ofs);
#else
    /* ofs might correspond to .addr_write, so use atomic_read */
    return atomic_read((target_ulong *)((uintptr_t)entry + ofs));
#endif
}

/* Return true if ADDR is present in the victim tlb, and has been copied
   back to the main tlb.  */
static bool victim_tlb_hit(CPUArchState *env, size_t mmu_idx, size_t index,
                           size_t elt_ofs, target_ulong page)
{
    size_t vidx;

    assert_cpu_is_self(env_cpu(env));
    for (vidx = 0; vidx < CPU_VTLB_SIZE; ++vidx) {
        CPUTLBEntry *vtlb = &env_tlb(env)->d[mmu_idx].vtable[vidx];
        target_ulong cmp;

        /* elt_ofs might correspond to .addr_write, so use atomic_read */
#if TCG_OVERSIZED_GUEST
        cmp = *(target_ulong *)((uintptr_t)vtlb + elt_ofs);
#else
        cmp = atomic_read((target_ulong *)((uintptr_t)vtlb + elt_ofs));
#endif

        if (cmp == page) {
            /* Found entry in victim tlb, swap tlb and iotlb.  */
            CPUTLBEntry tmptlb, *tlb = &env_tlb(env)->f[mmu_idx].table[index];

            qemu_spin_lock(&env_tlb(env)->c.lock);
            copy_tlb_helper_locked(&tmptlb, tlb);
            copy_tlb_helper_locked(tlb, vtlb);
            copy_tlb_helper_locked(vtlb, &tmptlb);
            qemu_spin_unlock(&env_tlb(env)->c.lock);

            CPUIOTLBEntry tmpio, *io = &env_tlb(env)->d[mmu_idx].iotlb[index];
            CPUIOTLBEntry *vio = &env_tlb(env)->d[mmu_idx].viotlb[vidx];
            tmpio = *io; *io = *vio; *vio = tmpio;
            return true;
        }
    }
    return false;
}

/* Macro to call the above, with local variables from the use context.  */
#define VICTIM_TLB_HIT(TY, ADDR) \
  victim_tlb_hit(env, mmu_idx, index, offsetof(CPUTLBEntry, TY), \
                 (ADDR) & TARGET_PAGE_MASK)

/*
 * Return a ram_addr_t for the virtual address for execution.
 *
 * Return -1 if we can't translate and execute from an entire page
 * of RAM.  This will force us to execute by loading and translating
 * one insn at a time, without caching.
 *
 * NOTE: This function will trigger an exception if the page is
 * not executable.
 */
tb_page_addr_t get_page_addr_code(CPUArchState *env, target_ulong addr)
{
    uintptr_t mmu_idx = cpu_mmu_index(env, true);
    uintptr_t index = tlb_index(env, mmu_idx, addr);
    CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
    void *p;

    if (unlikely(!tlb_hit(entry->addr_code, addr))) {
        if (!VICTIM_TLB_HIT(addr_code, addr)) {
            tlb_fill(env_cpu(env), addr, 0, MMU_INST_FETCH, mmu_idx, 0);
            index = tlb_index(env, mmu_idx, addr);
            entry = tlb_entry(env, mmu_idx, addr);

            if (unlikely(entry->addr_code & TLB_INVALID_MASK)) {
                /*
                 * The MMU protection covers a smaller range than a target
                 * page, so we must redo the MMU check for every insn.
                 */
                return -1;
            }
        }
        assert(tlb_hit(entry->addr_code, addr));
    }

    if (unlikely(entry->addr_code & TLB_MMIO)) {
        /* The region is not backed by RAM.  */
        return -1;
    }

    p = (void *)((uintptr_t)addr + entry->addend);
    return qemu_ram_addr_from_host_nofail(p);
}

static void notdirty_write(CPUState *cpu, vaddr mem_vaddr, unsigned size,
                           CPUIOTLBEntry *iotlbentry, uintptr_t retaddr)
{
    ram_addr_t ram_addr = mem_vaddr + iotlbentry->addr;

    trace_memory_notdirty_write_access(mem_vaddr, ram_addr, size);

    if (!cpu_physical_memory_get_dirty_flag(ram_addr, DIRTY_MEMORY_CODE)) {
        struct page_collection *pages
            = page_collection_lock(ram_addr, ram_addr + size);
        tb_invalidate_phys_page_fast(pages, ram_addr, size, retaddr);
        page_collection_unlock(pages);
    }

    /*
     * Set both VGA and migration bits for simplicity and to remove
     * the notdirty callback faster.
     */
    cpu_physical_memory_set_dirty_range(ram_addr, size, DIRTY_CLIENTS_NOCODE);

    /* We remove the notdirty callback only if the code has been flushed. */
    if (!cpu_physical_memory_is_clean(ram_addr)) {
        trace_memory_notdirty_set_dirty(mem_vaddr);
        tlb_set_dirty(cpu, mem_vaddr);
    }
}

/*
 * Probe for whether the specified guest access is permitted. If it is not
 * permitted then an exception will be taken in the same way as if this
 * were a real access (and we will not return).
 * If the size is 0 or the page requires I/O access, returns NULL; otherwise,
 * returns the address of the host page similar to tlb_vaddr_to_host().
 */
void *probe_access(CPUArchState *env, target_ulong addr, int size,
                   MMUAccessType access_type, int mmu_idx, uintptr_t retaddr)
{
    uintptr_t index = tlb_index(env, mmu_idx, addr);
    CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
    target_ulong tlb_addr;
    size_t elt_ofs;
    int wp_access;

    g_assert(-(addr | TARGET_PAGE_MASK) >= size);

    switch (access_type) {
    case MMU_DATA_LOAD:
        elt_ofs = offsetof(CPUTLBEntry, addr_read);
        wp_access = BP_MEM_READ;
        break;
    case MMU_DATA_STORE:
        elt_ofs = offsetof(CPUTLBEntry, addr_write);
        wp_access = BP_MEM_WRITE;
        break;
    case MMU_INST_FETCH:
        elt_ofs = offsetof(CPUTLBEntry, addr_code);
        wp_access = BP_MEM_READ;
        break;
    default:
        g_assert_not_reached();
    }
    tlb_addr = tlb_read_ofs(entry, elt_ofs);

    if (unlikely(!tlb_hit(tlb_addr, addr))) {
        if (!victim_tlb_hit(env, mmu_idx, index, elt_ofs,
                            addr & TARGET_PAGE_MASK)) {
            tlb_fill(env_cpu(env), addr, size, access_type, mmu_idx, retaddr);
            /* TLB resize via tlb_fill may have moved the entry. */
            index = tlb_index(env, mmu_idx, addr);
            entry = tlb_entry(env, mmu_idx, addr);
        }
        tlb_addr = tlb_read_ofs(entry, elt_ofs);
    }

    if (!size) {
        return NULL;
    }

    if (unlikely(tlb_addr & TLB_FLAGS_MASK)) {
        CPUIOTLBEntry *iotlbentry = &env_tlb(env)->d[mmu_idx].iotlb[index];

        /* Reject I/O access, or other required slow-path.  */
        if (tlb_addr & (TLB_MMIO | TLB_BSWAP | TLB_DISCARD_WRITE)) {
            return NULL;
        }

        /* Handle watchpoints.  */
        if (tlb_addr & TLB_WATCHPOINT) {
            cpu_check_watchpoint(env_cpu(env), addr, size,
                                 iotlbentry->attrs, wp_access, retaddr);
        }

        /* Handle clean RAM pages.  */
        if (tlb_addr & TLB_NOTDIRTY) {
            notdirty_write(env_cpu(env), addr, size, iotlbentry, retaddr);
        }
    }

    return (void *)((uintptr_t)addr + entry->addend);
}

void *tlb_vaddr_to_host(CPUArchState *env, abi_ptr addr,
                        MMUAccessType access_type, int mmu_idx)
{
    CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
    uintptr_t tlb_addr, page;
    size_t elt_ofs;

    switch (access_type) {
    case MMU_DATA_LOAD:
        elt_ofs = offsetof(CPUTLBEntry, addr_read);
        break;
    case MMU_DATA_STORE:
        elt_ofs = offsetof(CPUTLBEntry, addr_write);
        break;
    case MMU_INST_FETCH:
        elt_ofs = offsetof(CPUTLBEntry, addr_code);
        break;
    default:
        g_assert_not_reached();
    }

    page = addr & TARGET_PAGE_MASK;
    tlb_addr = tlb_read_ofs(entry, elt_ofs);

    if (!tlb_hit_page(tlb_addr, page)) {
        uintptr_t index = tlb_index(env, mmu_idx, addr);

        if (!victim_tlb_hit(env, mmu_idx, index, elt_ofs, page)) {
            CPUState *cs = env_cpu(env);
            CPUClass *cc = CPU_GET_CLASS(cs);

            if (!cc->tlb_fill(cs, addr, 0, access_type, mmu_idx, true, 0)) {
                /* Non-faulting page table read failed.  */
                return NULL;
            }

            /* TLB resize via tlb_fill may have moved the entry.  */
            entry = tlb_entry(env, mmu_idx, addr);
        }
        tlb_addr = tlb_read_ofs(entry, elt_ofs);
    }

    if (tlb_addr & ~TARGET_PAGE_MASK) {
        /* IO access */
        return NULL;
    }

    return (void *)((uintptr_t)addr + entry->addend);
}

/* Probe for a read-modify-write atomic operation.  Do not allow unaligned
 * operations, or io operations to proceed.  Return the host address.  */
static void *atomic_mmu_lookup(CPUArchState *env, target_ulong addr,
                               TCGMemOpIdx oi, uintptr_t retaddr)
{
    size_t mmu_idx = get_mmuidx(oi);
    uintptr_t index = tlb_index(env, mmu_idx, addr);
    CPUTLBEntry *tlbe = tlb_entry(env, mmu_idx, addr);
    target_ulong tlb_addr = tlb_addr_write(tlbe);
    MemOp mop = get_memop(oi);
    int a_bits = get_alignment_bits(mop);
    int s_bits = mop & MO_SIZE;
    void *hostaddr;

    /* Adjust the given return address.  */
    retaddr -= GETPC_ADJ;

    /* Enforce guest required alignment.  */
    if (unlikely(a_bits > 0 && (addr & ((1 << a_bits) - 1)))) {
        /* ??? Maybe indicate atomic op to cpu_unaligned_access */
        cpu_unaligned_access(env_cpu(env), addr, MMU_DATA_STORE,
                             mmu_idx, retaddr);
    }

    /* Enforce qemu required alignment.  */
    if (unlikely(addr & ((1 << s_bits) - 1))) {
        /* We get here if guest alignment was not requested,
           or was not enforced by cpu_unaligned_access above.
           We might widen the access and emulate, but for now
           mark an exception and exit the cpu loop.  */
        goto stop_the_world;
    }

    /* Check TLB entry and enforce page permissions.  */
    if (!tlb_hit(tlb_addr, addr)) {
        if (!VICTIM_TLB_HIT(addr_write, addr)) {
            tlb_fill(env_cpu(env), addr, 1 << s_bits, MMU_DATA_STORE,
                     mmu_idx, retaddr);
            index = tlb_index(env, mmu_idx, addr);
            tlbe = tlb_entry(env, mmu_idx, addr);
        }
        tlb_addr = tlb_addr_write(tlbe) & ~TLB_INVALID_MASK;
    }

    /* Notice an IO access or a needs-MMU-lookup access */
    if (unlikely(tlb_addr & TLB_MMIO)) {
        /* There's really nothing that can be done to
           support this apart from stop-the-world.  */
        goto stop_the_world;
    }

    /* Let the guest notice RMW on a write-only page.  */
    if (unlikely(tlbe->addr_read != (tlb_addr & ~TLB_NOTDIRTY))) {
        tlb_fill(env_cpu(env), addr, 1 << s_bits, MMU_DATA_LOAD,
                 mmu_idx, retaddr);
        /* Since we don't support reads and writes to different addresses,
           and we do have the proper page loaded for write, this shouldn't
           ever return.  But just in case, handle via stop-the-world.  */
        goto stop_the_world;
    }

    hostaddr = (void *)((uintptr_t)addr + tlbe->addend);

    if (unlikely(tlb_addr & TLB_NOTDIRTY)) {
        notdirty_write(env_cpu(env), addr, 1 << s_bits,
                       &env_tlb(env)->d[mmu_idx].iotlb[index], retaddr);
    }

    return hostaddr;

 stop_the_world:
    cpu_loop_exit_atomic(env_cpu(env), retaddr);
}

/*
 * Load Helpers
 *
 * We support two different access types. SOFTMMU_CODE_ACCESS is
 * specifically for reading instructions from system memory. It is
 * called by the translation loop and in some helpers where the code
 * is disassembled. It shouldn't be called directly by guest code.
 */

typedef uint64_t FullLoadHelper(CPUArchState *env, target_ulong addr,
                                TCGMemOpIdx oi, uintptr_t retaddr);

static inline uint64_t QEMU_ALWAYS_INLINE
load_memop(const void *haddr, MemOp op)
{
    switch (op) {
    case MO_UB:
        return ldub_p(haddr);
    case MO_BEUW:
        return lduw_be_p(haddr);
    case MO_LEUW:
        return lduw_le_p(haddr);
    case MO_BEUL:
        return (uint32_t)ldl_be_p(haddr);
    case MO_LEUL:
        return (uint32_t)ldl_le_p(haddr);
    case MO_BEQ:
        return ldq_be_p(haddr);
    case MO_LEQ:
        return ldq_le_p(haddr);
    default:
        qemu_build_not_reached();
    }
}

static inline uint64_t QEMU_ALWAYS_INLINE
load_helper(CPUArchState *env, target_ulong addr, TCGMemOpIdx oi,
            uintptr_t retaddr, MemOp op, bool code_read,
            FullLoadHelper *full_load)
{
    uintptr_t mmu_idx = get_mmuidx(oi);
    uintptr_t index = tlb_index(env, mmu_idx, addr);
    CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
    target_ulong tlb_addr = code_read ? entry->addr_code : entry->addr_read;
    const size_t tlb_off = code_read ?
        offsetof(CPUTLBEntry, addr_code) : offsetof(CPUTLBEntry, addr_read);
    const MMUAccessType access_type =
        code_read ? MMU_INST_FETCH : MMU_DATA_LOAD;
    unsigned a_bits = get_alignment_bits(get_memop(oi));
    void *haddr;
    uint64_t res;
    size_t size = memop_size(op);

    /* Handle CPU specific unaligned behaviour */
    if (addr & ((1 << a_bits) - 1)) {
        cpu_unaligned_access(env_cpu(env), addr, access_type,
                             mmu_idx, retaddr);
    }

    /* If the TLB entry is for a different page, reload and try again.  */
    if (!tlb_hit(tlb_addr, addr)) {
        if (!victim_tlb_hit(env, mmu_idx, index, tlb_off,
                            addr & TARGET_PAGE_MASK)) {
            tlb_fill(env_cpu(env), addr, size,
                     access_type, mmu_idx, retaddr);
            index = tlb_index(env, mmu_idx, addr);
            entry = tlb_entry(env, mmu_idx, addr);
        }
        tlb_addr = code_read ? entry->addr_code : entry->addr_read;
        tlb_addr &= ~TLB_INVALID_MASK;
    }

    /* Handle anything that isn't just a straight memory access.  */
    if (unlikely(tlb_addr & ~TARGET_PAGE_MASK)) {
        CPUIOTLBEntry *iotlbentry;
        bool need_swap;

        /* For anything that is unaligned, recurse through full_load.  */
        if ((addr & (size - 1)) != 0) {
            goto do_unaligned_access;
        }

        iotlbentry = &env_tlb(env)->d[mmu_idx].iotlb[index];

        /* Handle watchpoints.  */
        if (unlikely(tlb_addr & TLB_WATCHPOINT)) {
            /* On watchpoint hit, this will longjmp out.  */
            cpu_check_watchpoint(env_cpu(env), addr, size,
                                 iotlbentry->attrs, BP_MEM_READ, retaddr);
        }

        need_swap = size > 1 && (tlb_addr & TLB_BSWAP);

        /* Handle I/O access.  */
        if (likely(tlb_addr & TLB_MMIO)) {
            return io_readx(env, iotlbentry, mmu_idx, addr, retaddr,
                            access_type, op ^ (need_swap * MO_BSWAP));
        }

        haddr = (void *)((uintptr_t)addr + entry->addend);

        /*
         * Keep these two load_memop separate to ensure that the compiler
         * is able to fold the entire function to a single instruction.
         * There is a build-time assert inside to remind you of this.  ;-)
         */
        if (unlikely(need_swap)) {
            return load_memop(haddr, op ^ MO_BSWAP);
        }
        return load_memop(haddr, op);
    }

    /* Handle slow unaligned access (it spans two pages or IO).  */
    if (size > 1
        && unlikely((addr & ~TARGET_PAGE_MASK) + size - 1
                    >= TARGET_PAGE_SIZE)) {
        target_ulong addr1, addr2;
        uint64_t r1, r2;
        unsigned shift;
    do_unaligned_access:
        addr1 = addr & ~((target_ulong)size - 1);
        addr2 = addr1 + size;
        r1 = full_load(env, addr1, oi, retaddr);
        r2 = full_load(env, addr2, oi, retaddr);
        shift = (addr & (size - 1)) * 8;

        if (memop_big_endian(op)) {
            /* Big-endian combine.  */
            res = (r1 << shift) | (r2 >> ((size * 8) - shift));
        } else {
            /* Little-endian combine.  */
            res = (r1 >> shift) | (r2 << ((size * 8) - shift));
        }
        return res & MAKE_64BIT_MASK(0, size * 8);
    }

    haddr = (void *)((uintptr_t)addr + entry->addend);
    return load_memop(haddr, op);
}

/*
 * For the benefit of TCG generated code, we want to avoid the
 * complication of ABI-specific return type promotion and always
 * return a value extended to the register size of the host. This is
 * tcg_target_long, except in the case of a 32-bit host and 64-bit
 * data, and for that we always have uint64_t.
 *
 * We don't bother with this widened value for SOFTMMU_CODE_ACCESS.
 */

static uint64_t full_ldub_mmu(CPUArchState *env, target_ulong addr,
                              TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_UB, false, full_ldub_mmu);
}

tcg_target_ulong helper_ret_ldub_mmu(CPUArchState *env, target_ulong addr,
                                     TCGMemOpIdx oi, uintptr_t retaddr)
{
    return full_ldub_mmu(env, addr, oi, retaddr);
}

static uint64_t full_le_lduw_mmu(CPUArchState *env, target_ulong addr,
                                 TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_LEUW, false,
                       full_le_lduw_mmu);
}

tcg_target_ulong helper_le_lduw_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return full_le_lduw_mmu(env, addr, oi, retaddr);
}

static uint64_t full_be_lduw_mmu(CPUArchState *env, target_ulong addr,
                                 TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_BEUW, false,
                       full_be_lduw_mmu);
}

tcg_target_ulong helper_be_lduw_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return full_be_lduw_mmu(env, addr, oi, retaddr);
}

static uint64_t full_le_ldul_mmu(CPUArchState *env, target_ulong addr,
                                 TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_LEUL, false,
                       full_le_ldul_mmu);
}

tcg_target_ulong helper_le_ldul_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return full_le_ldul_mmu(env, addr, oi, retaddr);
}

static uint64_t full_be_ldul_mmu(CPUArchState *env, target_ulong addr,
                                 TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_BEUL, false,
                       full_be_ldul_mmu);
}

tcg_target_ulong helper_be_ldul_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return full_be_ldul_mmu(env, addr, oi, retaddr);
}

uint64_t helper_le_ldq_mmu(CPUArchState *env, target_ulong addr,
                           TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_LEQ, false,
                       helper_le_ldq_mmu);
}

uint64_t helper_be_ldq_mmu(CPUArchState *env, target_ulong addr,
                           TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_BEQ, false,
                       helper_be_ldq_mmu);
}

/*
 * Provide signed versions of the load routines as well.  We can of course
 * avoid this for 64-bit data, or for 32-bit data on 32-bit host.
 */


tcg_target_ulong helper_ret_ldsb_mmu(CPUArchState *env, target_ulong addr,
                                     TCGMemOpIdx oi, uintptr_t retaddr)
{
    return (int8_t)helper_ret_ldub_mmu(env, addr, oi, retaddr);
}

tcg_target_ulong helper_le_ldsw_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return (int16_t)helper_le_lduw_mmu(env, addr, oi, retaddr);
}

tcg_target_ulong helper_be_ldsw_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return (int16_t)helper_be_lduw_mmu(env, addr, oi, retaddr);
}

tcg_target_ulong helper_le_ldsl_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return (int32_t)helper_le_ldul_mmu(env, addr, oi, retaddr);
}

tcg_target_ulong helper_be_ldsl_mmu(CPUArchState *env, target_ulong addr,
                                    TCGMemOpIdx oi, uintptr_t retaddr)
{
    return (int32_t)helper_be_ldul_mmu(env, addr, oi, retaddr);
}

/*
 * Store Helpers
 */

static inline void QEMU_ALWAYS_INLINE
store_memop(void *haddr, uint64_t val, MemOp op)
{
    switch (op) {
    case MO_UB:
        stb_p(haddr, val);
        break;
    case MO_BEUW:
        stw_be_p(haddr, val);
        break;
    case MO_LEUW:
        stw_le_p(haddr, val);
        break;
    case MO_BEUL:
        stl_be_p(haddr, val);
        break;
    case MO_LEUL:
        stl_le_p(haddr, val);
        break;
    case MO_BEQ:
        stq_be_p(haddr, val);
        break;
    case MO_LEQ:
        stq_le_p(haddr, val);
        break;
    default:
        qemu_build_not_reached();
    }
}

static inline void QEMU_ALWAYS_INLINE
store_helper(CPUArchState *env, target_ulong addr, uint64_t val,
             TCGMemOpIdx oi, uintptr_t retaddr, MemOp op)
{
    uintptr_t mmu_idx = get_mmuidx(oi);
    uintptr_t index = tlb_index(env, mmu_idx, addr);
    CPUTLBEntry *entry = tlb_entry(env, mmu_idx, addr);
    target_ulong tlb_addr = tlb_addr_write(entry);
    const size_t tlb_off = offsetof(CPUTLBEntry, addr_write);
    unsigned a_bits = get_alignment_bits(get_memop(oi));
    void *haddr;
    size_t size = memop_size(op);

    /* Handle CPU specific unaligned behaviour */
    if (addr & ((1 << a_bits) - 1)) {
        cpu_unaligned_access(env_cpu(env), addr, MMU_DATA_STORE,
                             mmu_idx, retaddr);
    }

    /* If the TLB entry is for a different page, reload and try again.  */
    if (!tlb_hit(tlb_addr, addr)) {
        if (!victim_tlb_hit(env, mmu_idx, index, tlb_off,
            addr & TARGET_PAGE_MASK)) {
            tlb_fill(env_cpu(env), addr, size, MMU_DATA_STORE,
                     mmu_idx, retaddr);
            index = tlb_index(env, mmu_idx, addr);
            entry = tlb_entry(env, mmu_idx, addr);
        }
        tlb_addr = tlb_addr_write(entry) & ~TLB_INVALID_MASK;
    }

    /* Handle anything that isn't just a straight memory access.  */
    if (unlikely(tlb_addr & ~TARGET_PAGE_MASK)) {
        CPUIOTLBEntry *iotlbentry;
        bool need_swap;

        /* For anything that is unaligned, recurse through byte stores.  */
        if ((addr & (size - 1)) != 0) {
            goto do_unaligned_access;
        }

        iotlbentry = &env_tlb(env)->d[mmu_idx].iotlb[index];

        /* Handle watchpoints.  */
        if (unlikely(tlb_addr & TLB_WATCHPOINT)) {
            /* On watchpoint hit, this will longjmp out.  */
            cpu_check_watchpoint(env_cpu(env), addr, size,
                                 iotlbentry->attrs, BP_MEM_WRITE, retaddr);
        }

        need_swap = size > 1 && (tlb_addr & TLB_BSWAP);

        /* Handle I/O access.  */
        if (tlb_addr & TLB_MMIO) {
            io_writex(env, iotlbentry, mmu_idx, val, addr, retaddr,
                      op ^ (need_swap * MO_BSWAP));
            return;
        }

        /* Ignore writes to ROM.  */
        if (unlikely(tlb_addr & TLB_DISCARD_WRITE)) {
            return;
        }

        /* Handle clean RAM pages.  */
        if (tlb_addr & TLB_NOTDIRTY) {
            notdirty_write(env_cpu(env), addr, size, iotlbentry, retaddr);
        }

        haddr = (void *)((uintptr_t)addr + entry->addend);

        /*
         * Keep these two store_memop separate to ensure that the compiler
         * is able to fold the entire function to a single instruction.
         * There is a build-time assert inside to remind you of this.  ;-)
         */
        if (unlikely(need_swap)) {
            store_memop(haddr, val, op ^ MO_BSWAP);
        } else {
            store_memop(haddr, val, op);
        }
        return;
    }

    /* Handle slow unaligned access (it spans two pages or IO).  */
    if (size > 1
        && unlikely((addr & ~TARGET_PAGE_MASK) + size - 1
                     >= TARGET_PAGE_SIZE)) {
        int i;
        uintptr_t index2;
        CPUTLBEntry *entry2;
        target_ulong page2, tlb_addr2;
        size_t size2;

    do_unaligned_access:
        /*
         * Ensure the second page is in the TLB.  Note that the first page
         * is already guaranteed to be filled, and that the second page
         * cannot evict the first.
         */
        page2 = (addr + size) & TARGET_PAGE_MASK;
        size2 = (addr + size) & ~TARGET_PAGE_MASK;
        index2 = tlb_index(env, mmu_idx, page2);
        entry2 = tlb_entry(env, mmu_idx, page2);
        tlb_addr2 = tlb_addr_write(entry2);
        if (!tlb_hit_page(tlb_addr2, page2)) {
            if (!victim_tlb_hit(env, mmu_idx, index2, tlb_off, page2)) {
                tlb_fill(env_cpu(env), page2, size2, MMU_DATA_STORE,
                         mmu_idx, retaddr);
                index2 = tlb_index(env, mmu_idx, page2);
                entry2 = tlb_entry(env, mmu_idx, page2);
            }
            tlb_addr2 = tlb_addr_write(entry2);
        }

        /*
         * Handle watchpoints.  Since this may trap, all checks
         * must happen before any store.
         */
        if (unlikely(tlb_addr & TLB_WATCHPOINT)) {
            cpu_check_watchpoint(env_cpu(env), addr, size - size2,
                                 env_tlb(env)->d[mmu_idx].iotlb[index].attrs,
                                 BP_MEM_WRITE, retaddr);
        }
        if (unlikely(tlb_addr2 & TLB_WATCHPOINT)) {
            cpu_check_watchpoint(env_cpu(env), page2, size2,
                                 env_tlb(env)->d[mmu_idx].iotlb[index2].attrs,
                                 BP_MEM_WRITE, retaddr);
        }

        /*
         * XXX: not efficient, but simple.
         * This loop must go in the forward direction to avoid issues
         * with self-modifying code in Windows 64-bit.
         */
        for (i = 0; i < size; ++i) {
            uint8_t val8;
            if (memop_big_endian(op)) {
                /* Big-endian extract.  */
                val8 = val >> (((size - 1) * 8) - (i * 8));
            } else {
                /* Little-endian extract.  */
                val8 = val >> (i * 8);
            }
            helper_ret_stb_mmu(env, addr + i, val8, oi, retaddr);
        }
        return;
    }

    haddr = (void *)((uintptr_t)addr + entry->addend);
    store_memop(haddr, val, op);
}

void helper_ret_stb_mmu(CPUArchState *env, target_ulong addr, uint8_t val,
                        TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, MO_UB);
}

void helper_le_stw_mmu(CPUArchState *env, target_ulong addr, uint16_t val,
                       TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, MO_LEUW);
}

void helper_be_stw_mmu(CPUArchState *env, target_ulong addr, uint16_t val,
                       TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, MO_BEUW);
}

void helper_le_stl_mmu(CPUArchState *env, target_ulong addr, uint32_t val,
                       TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, MO_LEUL);
}

void helper_be_stl_mmu(CPUArchState *env, target_ulong addr, uint32_t val,
                       TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, MO_BEUL);
}

void helper_le_stq_mmu(CPUArchState *env, target_ulong addr, uint64_t val,
                       TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, MO_LEQ);
}

void helper_be_stq_mmu(CPUArchState *env, target_ulong addr, uint64_t val,
                       TCGMemOpIdx oi, uintptr_t retaddr)
{
    store_helper(env, addr, val, oi, retaddr, MO_BEQ);
}

/* First set of helpers allows passing in of OI and RETADDR.  This makes
   them callable from other helpers.  */

#define EXTRA_ARGS     , TCGMemOpIdx oi, uintptr_t retaddr
#define ATOMIC_NAME(X) \
    HELPER(glue(glue(glue(atomic_ ## X, SUFFIX), END), _mmu))
#define ATOMIC_MMU_DECLS
#define ATOMIC_MMU_LOOKUP atomic_mmu_lookup(env, addr, oi, retaddr)
#define ATOMIC_MMU_CLEANUP

#define DATA_SIZE 1
#include "atomic_template.h"

#define DATA_SIZE 2
#include "atomic_template.h"

#define DATA_SIZE 4
#include "atomic_template.h"

#ifdef CONFIG_ATOMIC64
#define DATA_SIZE 8
#include "atomic_template.h"
#endif

#if HAVE_CMPXCHG128 || HAVE_ATOMIC128
#define DATA_SIZE 16
#include "atomic_template.h"
#endif

/* Second set of helpers are directly callable from TCG as helpers.  */

#undef EXTRA_ARGS
#undef ATOMIC_NAME
#undef ATOMIC_MMU_LOOKUP
#define EXTRA_ARGS         , TCGMemOpIdx oi
#define ATOMIC_NAME(X)     HELPER(glue(glue(atomic_ ## X, SUFFIX), END))
#define ATOMIC_MMU_LOOKUP  atomic_mmu_lookup(env, addr, oi, GETPC())

#define DATA_SIZE 1
#include "atomic_template.h"

#define DATA_SIZE 2
#include "atomic_template.h"

#define DATA_SIZE 4
#include "atomic_template.h"

#ifdef CONFIG_ATOMIC64
#define DATA_SIZE 8
#include "atomic_template.h"
#endif

/* Code access functions.  */

static uint64_t full_ldub_cmmu(CPUArchState *env, target_ulong addr,
                               TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_8, true, full_ldub_cmmu);
}

uint8_t helper_ret_ldb_cmmu(CPUArchState *env, target_ulong addr,
                            TCGMemOpIdx oi, uintptr_t retaddr)
{
    return full_ldub_cmmu(env, addr, oi, retaddr);
}

static uint64_t full_le_lduw_cmmu(CPUArchState *env, target_ulong addr,
                                  TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_LEUW, true,
                       full_le_lduw_cmmu);
}

uint16_t helper_le_ldw_cmmu(CPUArchState *env, target_ulong addr,
                            TCGMemOpIdx oi, uintptr_t retaddr)
{
    return full_le_lduw_cmmu(env, addr, oi, retaddr);
}

static uint64_t full_be_lduw_cmmu(CPUArchState *env, target_ulong addr,
                                  TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_BEUW, true,
                       full_be_lduw_cmmu);
}

uint16_t helper_be_ldw_cmmu(CPUArchState *env, target_ulong addr,
                            TCGMemOpIdx oi, uintptr_t retaddr)
{
    return full_be_lduw_cmmu(env, addr, oi, retaddr);
}

static uint64_t full_le_ldul_cmmu(CPUArchState *env, target_ulong addr,
                                  TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_LEUL, true,
                       full_le_ldul_cmmu);
}

uint32_t helper_le_ldl_cmmu(CPUArchState *env, target_ulong addr,
                            TCGMemOpIdx oi, uintptr_t retaddr)
{
    return full_le_ldul_cmmu(env, addr, oi, retaddr);
}

static uint64_t full_be_ldul_cmmu(CPUArchState *env, target_ulong addr,
                                  TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_BEUL, true,
                       full_be_ldul_cmmu);
}

uint32_t helper_be_ldl_cmmu(CPUArchState *env, target_ulong addr,
                            TCGMemOpIdx oi, uintptr_t retaddr)
{
    return full_be_ldul_cmmu(env, addr, oi, retaddr);
}

uint64_t helper_le_ldq_cmmu(CPUArchState *env, target_ulong addr,
                            TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_LEQ, true,
                       helper_le_ldq_cmmu);
}

uint64_t helper_be_ldq_cmmu(CPUArchState *env, target_ulong addr,
                            TCGMemOpIdx oi, uintptr_t retaddr)
{
    return load_helper(env, addr, oi, retaddr, MO_BEQ, true,
                       helper_be_ldq_cmmu);
}