summaryrefslogtreecommitdiffstats
path: root/arch/tile/kernel/intvec_32.S
blob: f3d26f48e659aa72dcf091683e4489afec7bda04 (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
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
/*
 * Copyright 2010 Tilera Corporation. All Rights Reserved.
 *
 *   This program is free software; you can redistribute it and/or
 *   modify it under the terms of the GNU General Public License
 *   as published by the Free Software Foundation, version 2.
 *
 *   This program 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, GOOD TITLE or
 *   NON INFRINGEMENT.  See the GNU General Public License for
 *   more details.
 *
 * Linux interrupt vectors.
 */

#include <linux/linkage.h>
#include <linux/errno.h>
#include <linux/init.h>
#include <linux/unistd.h>
#include <asm/ptrace.h>
#include <asm/thread_info.h>
#include <asm/irqflags.h>
#include <asm/atomic_32.h>
#include <asm/asm-offsets.h>
#include <hv/hypervisor.h>
#include <arch/abi.h>
#include <arch/interrupts.h>
#include <arch/spr_def.h>

#define PTREGS_PTR(reg, ptreg) addli reg, sp, C_ABI_SAVE_AREA_SIZE + (ptreg)

#define PTREGS_OFFSET_SYSCALL PTREGS_OFFSET_REG(TREG_SYSCALL_NR)

#if !CHIP_HAS_WH64()
	/* By making this an empty macro, we can use wh64 in the code. */
	.macro  wh64 reg
	.endm
#endif

	.macro  push_reg reg, ptr=sp, delta=-4
	{
	 sw     \ptr, \reg
	 addli  \ptr, \ptr, \delta
	}
	.endm

	.macro  pop_reg reg, ptr=sp, delta=4
	{
	 lw     \reg, \ptr
	 addli  \ptr, \ptr, \delta
	}
	.endm

	.macro  pop_reg_zero reg, zreg, ptr=sp, delta=4
	{
	 move   \zreg, zero
	 lw     \reg, \ptr
	 addi   \ptr, \ptr, \delta
	}
	.endm

	.macro  push_extra_callee_saves reg
	PTREGS_PTR(\reg, PTREGS_OFFSET_REG(51))
	push_reg r51, \reg
	push_reg r50, \reg
	push_reg r49, \reg
	push_reg r48, \reg
	push_reg r47, \reg
	push_reg r46, \reg
	push_reg r45, \reg
	push_reg r44, \reg
	push_reg r43, \reg
	push_reg r42, \reg
	push_reg r41, \reg
	push_reg r40, \reg
	push_reg r39, \reg
	push_reg r38, \reg
	push_reg r37, \reg
	push_reg r36, \reg
	push_reg r35, \reg
	push_reg r34, \reg, PTREGS_OFFSET_BASE - PTREGS_OFFSET_REG(34)
	.endm

	.macro  panic str
	.pushsection .rodata, "a"
1:
	.asciz  "\str"
	.popsection
	{
	 moveli r0, lo16(1b)
	}
	{
	 auli   r0, r0, ha16(1b)
	 jal    panic
	}
	.endm

#ifdef __COLLECT_LINKER_FEEDBACK__
	.pushsection .text.intvec_feedback,"ax"
intvec_feedback:
	.popsection
#endif

	/*
	 * Default interrupt handler.
	 *
	 * vecnum is where we'll put this code.
	 * c_routine is the C routine we'll call.
	 *
	 * The C routine is passed two arguments:
	 * - A pointer to the pt_regs state.
	 * - The interrupt vector number.
	 *
	 * The "processing" argument specifies the code for processing
	 * the interrupt. Defaults to "handle_interrupt".
	 */
	.macro  int_hand vecnum, vecname, c_routine, processing=handle_interrupt
	.org    (\vecnum << 8)
intvec_\vecname:
	.ifc    \vecnum, INT_SWINT_1
	blz     TREG_SYSCALL_NR_NAME, sys_cmpxchg
	.endif

	/* Temporarily save a register so we have somewhere to work. */

	mtspr   SPR_SYSTEM_SAVE_K_1, r0
	mfspr   r0, SPR_EX_CONTEXT_K_1

	/* The cmpxchg code clears sp to force us to reset it here on fault. */
	{
	 bz     sp, 2f
	 andi   r0, r0, SPR_EX_CONTEXT_1_1__PL_MASK  /* mask off ICS */
	}

	.ifc    \vecnum, INT_DOUBLE_FAULT
	/*
	 * For double-faults from user-space, fall through to the normal
	 * register save and stack setup path.  Otherwise, it's the
	 * hypervisor giving us one last chance to dump diagnostics, and we
	 * branch to the kernel_double_fault routine to do so.
	 */
	bz      r0, 1f
	j       _kernel_double_fault
1:
	.else
	/*
	 * If we're coming from user-space, then set sp to the top of
	 * the kernel stack.  Otherwise, assume sp is already valid.
	 */
	{
	 bnz    r0, 0f
	 move   r0, sp
	}
	.endif

	.ifc    \c_routine, do_page_fault
	/*
	 * The page_fault handler may be downcalled directly by the
	 * hypervisor even when Linux is running and has ICS set.
	 *
	 * In this case the contents of EX_CONTEXT_K_1 reflect the
	 * previous fault and can't be relied on to choose whether or
	 * not to reinitialize the stack pointer.  So we add a test
	 * to see whether SYSTEM_SAVE_K_2 has the high bit set,
	 * and if so we don't reinitialize sp, since we must be coming
	 * from Linux.  (In fact the precise case is !(val & ~1),
	 * but any Linux PC has to have the high bit set.)
	 *
	 * Note that the hypervisor *always* sets SYSTEM_SAVE_K_2 for
	 * any path that turns into a downcall to one of our TLB handlers.
	 */
	mfspr   r0, SPR_SYSTEM_SAVE_K_2
	{
	 blz    r0, 0f    /* high bit in S_S_1_2 is for a PC to use */
	 move   r0, sp
	}
	.endif

2:
	/*
	 * SYSTEM_SAVE_K_0 holds the cpu number in the low bits, and
	 * the current stack top in the higher bits.  So we recover
	 * our stack top by just masking off the low bits, then
	 * point sp at the top aligned address on the actual stack page.
	 */
	mfspr   r0, SPR_SYSTEM_SAVE_K_0
	mm      r0, r0, zero, LOG2_NR_CPU_IDS, 31

0:
	/*
	 * Align the stack mod 64 so we can properly predict what
	 * cache lines we need to write-hint to reduce memory fetch
	 * latency as we enter the kernel.  The layout of memory is
	 * as follows, with cache line 0 at the lowest VA, and cache
	 * line 4 just below the r0 value this "andi" computes.
	 * Note that we never write to cache line 4, and we skip
	 * cache line 1 for syscalls.
	 *
	 *    cache line 4: ptregs padding (two words)
	 *    cache line 3: r46...lr, pc, ex1, faultnum, orig_r0, flags, pad
	 *    cache line 2: r30...r45
	 *    cache line 1: r14...r29
	 *    cache line 0: 2 x frame, r0..r13
	 */
#if STACK_TOP_DELTA != 64
#error STACK_TOP_DELTA must be 64 for assumptions here and in task_pt_regs()
#endif
	andi    r0, r0, -64

	/*
	 * Push the first four registers on the stack, so that we can set
	 * them to vector-unique values before we jump to the common code.
	 *
	 * Registers are pushed on the stack as a struct pt_regs,
	 * with the sp initially just above the struct, and when we're
	 * done, sp points to the base of the struct, minus
	 * C_ABI_SAVE_AREA_SIZE, so we can directly jal to C code.
	 *
	 * This routine saves just the first four registers, plus the
	 * stack context so we can do proper backtracing right away,
	 * and defers to handle_interrupt to save the rest.
	 * The backtracer needs pc, ex1, lr, sp, r52, and faultnum.
	 */
	addli   r0, r0, PTREGS_OFFSET_LR - (PTREGS_SIZE + KSTK_PTREGS_GAP)
	wh64    r0    /* cache line 3 */
	{
	 sw     r0, lr
	 addli  r0, r0, PTREGS_OFFSET_SP - PTREGS_OFFSET_LR
	}
	{
	 sw     r0, sp
	 addli  sp, r0, PTREGS_OFFSET_REG(52) - PTREGS_OFFSET_SP
	}
	{
	 sw     sp, r52
	 addli  sp, sp, PTREGS_OFFSET_REG(1) - PTREGS_OFFSET_REG(52)
	}
	wh64    sp    /* cache line 0 */
	{
	 sw     sp, r1
	 addli  sp, sp, PTREGS_OFFSET_REG(2) - PTREGS_OFFSET_REG(1)
	}
	{
	 sw     sp, r2
	 addli  sp, sp, PTREGS_OFFSET_REG(3) - PTREGS_OFFSET_REG(2)
	}
	{
	 sw     sp, r3
	 addli  sp, sp, PTREGS_OFFSET_PC - PTREGS_OFFSET_REG(3)
	}
	mfspr   r0, SPR_EX_CONTEXT_K_0
	.ifc \processing,handle_syscall
	/*
	 * Bump the saved PC by one bundle so that when we return, we won't
	 * execute the same swint instruction again.  We need to do this while
	 * we're in the critical section.
	 */
	addi    r0, r0, 8
	.endif
	{
	 sw     sp, r0
	 addli  sp, sp, PTREGS_OFFSET_EX1 - PTREGS_OFFSET_PC
	}
	mfspr   r0, SPR_EX_CONTEXT_K_1
	{
	 sw     sp, r0
	 addi   sp, sp, PTREGS_OFFSET_FAULTNUM - PTREGS_OFFSET_EX1
	/*
	 * Use r0 for syscalls so it's a temporary; use r1 for interrupts
	 * so that it gets passed through unchanged to the handler routine.
	 * Note that the .if conditional confusingly spans bundles.
	 */
	 .ifc \processing,handle_syscall
	 movei  r0, \vecnum
	}
	{
	 sw     sp, r0
	 .else
	 movei  r1, \vecnum
	}
	{
	 sw     sp, r1
	 .endif
	 addli  sp, sp, PTREGS_OFFSET_REG(0) - PTREGS_OFFSET_FAULTNUM
	}
	mfspr   r0, SPR_SYSTEM_SAVE_K_1    /* Original r0 */
	{
	 sw     sp, r0
	 addi   sp, sp, -PTREGS_OFFSET_REG(0) - 4
	}
	{
	 sw     sp, zero        /* write zero into "Next SP" frame pointer */
	 addi   sp, sp, -4      /* leave SP pointing at bottom of frame */
	}
	.ifc \processing,handle_syscall
	j       handle_syscall
	.else
	/*
	 * Capture per-interrupt SPR context to registers.
	 * We overload the meaning of r3 on this path such that if its bit 31
	 * is set, we have to mask all interrupts including NMIs before
	 * clearing the interrupt critical section bit.
	 * See discussion below at "finish_interrupt_save".
	 */
	.ifc \c_routine, do_page_fault
	mfspr   r2, SPR_SYSTEM_SAVE_K_3   /* address of page fault */
	mfspr   r3, SPR_SYSTEM_SAVE_K_2   /* info about page fault */
	.else
	.ifc \vecnum, INT_DOUBLE_FAULT
	{
	 mfspr  r2, SPR_SYSTEM_SAVE_K_2   /* double fault info from HV */
	 movei  r3, 0
	}
	.else
	.ifc \c_routine, do_trap
	{
	 mfspr  r2, GPV_REASON
	 movei  r3, 0
	}
	.else
	.ifc \c_routine, op_handle_perf_interrupt
	{
	 mfspr  r2, PERF_COUNT_STS
	 movei  r3, -1   /* not used, but set for consistency */
	}
	.else
#if CHIP_HAS_AUX_PERF_COUNTERS()
	.ifc \c_routine, op_handle_aux_perf_interrupt
	{
	 mfspr  r2, AUX_PERF_COUNT_STS
	 movei  r3, -1   /* not used, but set for consistency */
	}
	.else
#endif
	movei   r3, 0
#if CHIP_HAS_AUX_PERF_COUNTERS()
	.endif
#endif
	.endif
	.endif
	.endif
	.endif
	/* Put function pointer in r0 */
	moveli  r0, lo16(\c_routine)
	{
	 auli   r0, r0, ha16(\c_routine)
	 j       \processing
	}
	.endif
	ENDPROC(intvec_\vecname)

#ifdef __COLLECT_LINKER_FEEDBACK__
	.pushsection .text.intvec_feedback,"ax"
	.org    (\vecnum << 5)
	FEEDBACK_ENTER_EXPLICIT(intvec_\vecname, .intrpt1, 1 << 8)
	jrp     lr
	.popsection
#endif

	.endm


	/*
	 * Save the rest of the registers that we didn't save in the actual
	 * vector itself.  We can't use r0-r10 inclusive here.
	 */
	.macro  finish_interrupt_save, function

	/* If it's a syscall, save a proper orig_r0, otherwise just zero. */
	PTREGS_PTR(r52, PTREGS_OFFSET_ORIG_R0)
	{
	 .ifc \function,handle_syscall
	 sw     r52, r0
	 .else
	 sw     r52, zero
	 .endif
	 PTREGS_PTR(r52, PTREGS_OFFSET_TP)
	}

	/*
	 * For ordinary syscalls, we save neither caller- nor callee-
	 * save registers, since the syscall invoker doesn't expect the
	 * caller-saves to be saved, and the called kernel functions will
	 * take care of saving the callee-saves for us.
	 *
	 * For interrupts we save just the caller-save registers.  Saving
	 * them is required (since the "caller" can't save them).  Again,
	 * the called kernel functions will restore the callee-save
	 * registers for us appropriately.
	 *
	 * On return, we normally restore nothing special for syscalls,
	 * and just the caller-save registers for interrupts.
	 *
	 * However, there are some important caveats to all this:
	 *
	 * - We always save a few callee-save registers to give us
	 *   some scratchpad registers to carry across function calls.
	 *
	 * - fork/vfork/etc require us to save all the callee-save
	 *   registers, which we do in PTREGS_SYSCALL_ALL_REGS, below.
	 *
	 * - We always save r0..r5 and r10 for syscalls, since we need
	 *   to reload them a bit later for the actual kernel call, and
	 *   since we might need them for -ERESTARTNOINTR, etc.
	 *
	 * - Before invoking a signal handler, we save the unsaved
	 *   callee-save registers so they are visible to the
	 *   signal handler or any ptracer.
	 *
	 * - If the unsaved callee-save registers are modified, we set
	 *   a bit in pt_regs so we know to reload them from pt_regs
	 *   and not just rely on the kernel function unwinding.
	 *   (Done for ptrace register writes and SA_SIGINFO handler.)
	 */
	{
	 sw     r52, tp
	 PTREGS_PTR(r52, PTREGS_OFFSET_REG(33))
	}
	wh64    r52    /* cache line 2 */
	push_reg r33, r52
	push_reg r32, r52
	push_reg r31, r52
	.ifc \function,handle_syscall
	push_reg r30, r52, PTREGS_OFFSET_SYSCALL - PTREGS_OFFSET_REG(30)
	push_reg TREG_SYSCALL_NR_NAME, r52, \
	  PTREGS_OFFSET_REG(5) - PTREGS_OFFSET_SYSCALL
	.else

	push_reg r30, r52, PTREGS_OFFSET_REG(29) - PTREGS_OFFSET_REG(30)
	wh64    r52    /* cache line 1 */
	push_reg r29, r52
	push_reg r28, r52
	push_reg r27, r52
	push_reg r26, r52
	push_reg r25, r52
	push_reg r24, r52
	push_reg r23, r52
	push_reg r22, r52
	push_reg r21, r52
	push_reg r20, r52
	push_reg r19, r52
	push_reg r18, r52
	push_reg r17, r52
	push_reg r16, r52
	push_reg r15, r52
	push_reg r14, r52
	push_reg r13, r52
	push_reg r12, r52
	push_reg r11, r52
	push_reg r10, r52
	push_reg r9, r52
	push_reg r8, r52
	push_reg r7, r52
	push_reg r6, r52

	.endif

	push_reg r5, r52
	sw      r52, r4

	/* Load tp with our per-cpu offset. */
#ifdef CONFIG_SMP
	{
	 mfspr  r20, SPR_SYSTEM_SAVE_K_0
	 moveli r21, lo16(__per_cpu_offset)
	}
	{
	 auli   r21, r21, ha16(__per_cpu_offset)
	 mm     r20, r20, zero, 0, LOG2_NR_CPU_IDS-1
	}
	s2a     r20, r20, r21
	lw      tp, r20
#else
	move    tp, zero
#endif

	/*
	 * If we will be returning to the kernel, we will need to
	 * reset the interrupt masks to the state they had before.
	 * Set DISABLE_IRQ in flags iff we came from PL1 with irqs disabled.
	 * We load flags in r32 here so we can jump to .Lrestore_regs
	 * directly after do_page_fault_ics() if necessary.
	 */
	mfspr   r32, SPR_EX_CONTEXT_K_1
	{
	 andi   r32, r32, SPR_EX_CONTEXT_1_1__PL_MASK  /* mask off ICS */
	 PTREGS_PTR(r21, PTREGS_OFFSET_FLAGS)
	}
	bzt     r32, 1f       /* zero if from user space */
	IRQS_DISABLED(r32)    /* zero if irqs enabled */
#if PT_FLAGS_DISABLE_IRQ != 1
# error Value of IRQS_DISABLED used to set PT_FLAGS_DISABLE_IRQ; fix
#endif
1:
	.ifnc \function,handle_syscall
	/* Record the fact that we saved the caller-save registers above. */
	ori     r32, r32, PT_FLAGS_CALLER_SAVES
	.endif
	sw      r21, r32

#ifdef __COLLECT_LINKER_FEEDBACK__
	/*
	 * Notify the feedback routines that we were in the
	 * appropriate fixed interrupt vector area.  Note that we
	 * still have ICS set at this point, so we can't invoke any
	 * atomic operations or we will panic.  The feedback
	 * routines internally preserve r0..r10 and r30 up.
	 */
	.ifnc \function,handle_syscall
	shli    r20, r1, 5
	.else
	moveli  r20, INT_SWINT_1 << 5
	.endif
	addli   r20, r20, lo16(intvec_feedback)
	auli    r20, r20, ha16(intvec_feedback)
	jalr    r20

	/* And now notify the feedback routines that we are here. */
	FEEDBACK_ENTER(\function)
#endif

	/*
	 * we've captured enough state to the stack (including in
	 * particular our EX_CONTEXT state) that we can now release
	 * the interrupt critical section and replace it with our
	 * standard "interrupts disabled" mask value.  This allows
	 * synchronous interrupts (and profile interrupts) to punch
	 * through from this point onwards.
	 *
	 * If bit 31 of r3 is set during a non-NMI interrupt, we know we
	 * are on the path where the hypervisor has punched through our
	 * ICS with a page fault, so we call out to do_page_fault_ics()
	 * to figure out what to do with it.  If the fault was in
	 * an atomic op, we unlock the atomic lock, adjust the
	 * saved register state a little, and return "zero" in r4,
	 * falling through into the normal page-fault interrupt code.
	 * If the fault was in a kernel-space atomic operation, then
	 * do_page_fault_ics() resolves it itself, returns "one" in r4,
	 * and as a result goes directly to restoring registers and iret,
	 * without trying to adjust the interrupt masks at all.
	 * The do_page_fault_ics() API involves passing and returning
	 * a five-word struct (in registers) to avoid writing the
	 * save and restore code here.
	 */
	.ifc \function,handle_nmi
	IRQ_DISABLE_ALL(r20)
	.else
	.ifnc \function,handle_syscall
	bgezt   r3, 1f
	{
	 PTREGS_PTR(r0, PTREGS_OFFSET_BASE)
	 jal    do_page_fault_ics
	}
	FEEDBACK_REENTER(\function)
	bzt     r4, 1f
	j       .Lrestore_regs
1:
	.endif
	IRQ_DISABLE(r20, r21)
	.endif
	mtspr   INTERRUPT_CRITICAL_SECTION, zero

#if CHIP_HAS_WH64()
	/*
	 * Prepare the first 256 stack bytes to be rapidly accessible
	 * without having to fetch the background data.  We don't really
	 * know how far to write-hint, but kernel stacks generally
	 * aren't that big, and write-hinting here does take some time.
	 */
	addi    r52, sp, -64
	{
	 wh64   r52
	 addi   r52, r52, -64
	}
	{
	 wh64   r52
	 addi   r52, r52, -64
	}
	{
	 wh64   r52
	 addi   r52, r52, -64
	}
	wh64    r52
#endif

#ifdef CONFIG_TRACE_IRQFLAGS
	.ifnc \function,handle_nmi
	/*
	 * We finally have enough state set up to notify the irq
	 * tracing code that irqs were disabled on entry to the handler.
	 * The TRACE_IRQS_OFF call clobbers registers r0-r29.
	 * For syscalls, we already have the register state saved away
	 * on the stack, so we don't bother to do any register saves here,
	 * and later we pop the registers back off the kernel stack.
	 * For interrupt handlers, save r0-r3 in callee-saved registers.
	 */
	.ifnc \function,handle_syscall
	{ move r30, r0; move r31, r1 }
	{ move r32, r2; move r33, r3 }
	.endif
	TRACE_IRQS_OFF
	.ifnc \function,handle_syscall
	{ move r0, r30; move r1, r31 }
	{ move r2, r32; move r3, r33 }
	.endif
	.endif
#endif

	.endm

	.macro  check_single_stepping, kind, not_single_stepping
	/*
	 * Check for single stepping in user-level priv
	 *   kind can be "normal", "ill", or "syscall"
	 * At end, if fall-thru
	 *   r29: thread_info->step_state
	 *   r28: &pt_regs->pc
	 *   r27: pt_regs->pc
	 *   r26: thread_info->step_state->buffer
	 */

	/* Check for single stepping */
	GET_THREAD_INFO(r29)
	{
	 /* Get pointer to field holding step state */
	 addi   r29, r29, THREAD_INFO_STEP_STATE_OFFSET

	 /* Get pointer to EX1 in register state */
	 PTREGS_PTR(r27, PTREGS_OFFSET_EX1)
	}
	{
	 /* Get pointer to field holding PC */
	 PTREGS_PTR(r28, PTREGS_OFFSET_PC)

	 /* Load the pointer to the step state */
	 lw     r29, r29
	}
	/* Load EX1 */
	lw      r27, r27
	{
	 /* Points to flags */
	 addi   r23, r29, SINGLESTEP_STATE_FLAGS_OFFSET

	 /* No single stepping if there is no step state structure */
	 bzt    r29, \not_single_stepping
	}
	{
	 /* mask off ICS and any other high bits */
	 andi   r27, r27, SPR_EX_CONTEXT_1_1__PL_MASK

	 /* Load pointer to single step instruction buffer */
	 lw     r26, r29
	}
	/* Check priv state */
	bnz     r27, \not_single_stepping

	/* Get flags */
	lw      r22, r23
	{
	 /* Branch if single-step mode not enabled */
	 bbnst  r22, \not_single_stepping

	 /* Clear enabled flag */
	 andi   r22, r22, ~SINGLESTEP_STATE_MASK_IS_ENABLED
	}
	.ifc \kind,normal
	{
	 /* Load PC */
	 lw     r27, r28

	 /* Point to the entry containing the original PC */
	 addi   r24, r29, SINGLESTEP_STATE_ORIG_PC_OFFSET
	}
	{
	 /* Disable single stepping flag */
	 sw     r23, r22
	}
	{
	 /* Get the original pc */
	 lw     r24, r24

	 /* See if the PC is at the start of the single step buffer */
	 seq    r25, r26, r27
	}
	/*
	 * NOTE: it is really expected that the PC be in the single step buffer
	 *       at this point
	 */
	bzt     r25, \not_single_stepping

	/* Restore the original PC */
	sw      r28, r24
	.else
	.ifc \kind,syscall
	{
	 /* Load PC */
	 lw     r27, r28

	 /* Point to the entry containing the next PC */
	 addi   r24, r29, SINGLESTEP_STATE_NEXT_PC_OFFSET
	}
	{
	 /* Increment the stopped PC by the bundle size */
	 addi   r26, r26, 8

	 /* Disable single stepping flag */
	 sw     r23, r22
	}
	{
	 /* Get the next pc */
	 lw     r24, r24

	 /*
	  * See if the PC is one bundle past the start of the
	  * single step buffer
	  */
	 seq    r25, r26, r27
	}
	{
	 /*
	  * NOTE: it is really expected that the PC be in the
	  * single step buffer at this point
	  */
	 bzt    r25, \not_single_stepping
	}
	/* Set to the next PC */
	sw      r28, r24
	.else
	{
	 /* Point to 3rd bundle in buffer */
	 addi   r25, r26, 16

	 /* Load PC */
	 lw      r27, r28
	}
	{
	 /* Disable single stepping flag */
	 sw      r23, r22

	 /* See if the PC is in the single step buffer */
	 slte_u  r24, r26, r27
	}
	{
	 slte_u r25, r27, r25

	 /*
	  * NOTE: it is really expected that the PC be in the
	  * single step buffer at this point
	  */
	 bzt    r24, \not_single_stepping
	}
	bzt     r25, \not_single_stepping
	.endif
	.endif
	.endm

	/*
	 * Redispatch a downcall.
	 */
	.macro  dc_dispatch vecnum, vecname
	.org    (\vecnum << 8)
intvec_\vecname:
	j       _hv_downcall_dispatch
	ENDPROC(intvec_\vecname)
	.endm

	/*
	 * Common code for most interrupts.  The C function we're eventually
	 * going to is in r0, and the faultnum is in r1; the original
	 * values for those registers are on the stack.
	 */
	.pushsection .text.handle_interrupt,"ax"
handle_interrupt:
	finish_interrupt_save handle_interrupt

	/*
	 * Check for if we are single stepping in user level. If so, then
	 * we need to restore the PC.
	 */

	check_single_stepping normal, .Ldispatch_interrupt
.Ldispatch_interrupt:

	/* Jump to the C routine; it should enable irqs as soon as possible. */
	{
	 jalr   r0
	 PTREGS_PTR(r0, PTREGS_OFFSET_BASE)
	}
	FEEDBACK_REENTER(handle_interrupt)
	{
	 movei  r30, 0   /* not an NMI */
	 j      interrupt_return
	}
	STD_ENDPROC(handle_interrupt)

/*
 * This routine takes a boolean in r30 indicating if this is an NMI.
 * If so, we also expect a boolean in r31 indicating whether to
 * re-enable the oprofile interrupts.
 *
 * Note that .Lresume_userspace is jumped to directly in several
 * places, and we need to make sure r30 is set correctly in those
 * callers as well.
 */
STD_ENTRY(interrupt_return)
	/* If we're resuming to kernel space, don't check thread flags. */
	{
	 bnz    r30, .Lrestore_all  /* NMIs don't special-case user-space */
	 PTREGS_PTR(r29, PTREGS_OFFSET_EX1)
	}
	lw      r29, r29
	andi    r29, r29, SPR_EX_CONTEXT_1_1__PL_MASK  /* mask off ICS */
	bzt     r29, .Lresume_userspace

#ifdef CONFIG_PREEMPT
	/* Returning to kernel space. Check if we need preemption. */
	GET_THREAD_INFO(r29)
	addli   r28, r29, THREAD_INFO_FLAGS_OFFSET
	{
	 lw     r28, r28
	 addli  r29, r29, THREAD_INFO_PREEMPT_COUNT_OFFSET
	}
	{
	 andi   r28, r28, _TIF_NEED_RESCHED
	 lw     r29, r29
	}
	bzt     r28, 1f
	bnz     r29, 1f
	jal     preempt_schedule_irq
	FEEDBACK_REENTER(interrupt_return)
1:
#endif

	/* If we're resuming to _cpu_idle_nap, bump PC forward by 8. */
	{
	 PTREGS_PTR(r29, PTREGS_OFFSET_PC)
	 moveli r27, lo16(_cpu_idle_nap)
	}
	{
	 lw     r28, r29
	 auli   r27, r27, ha16(_cpu_idle_nap)
	}
	{
	 seq    r27, r27, r28
	}
	{
	 bbns   r27, .Lrestore_all
	 addi   r28, r28, 8
	}
	sw      r29, r28
	j       .Lrestore_all

.Lresume_userspace:
	FEEDBACK_REENTER(interrupt_return)

	/*
	 * Use r33 to hold whether we have already loaded the callee-saves
	 * into ptregs.  We don't want to do it twice in this loop, since
	 * then we'd clobber whatever changes are made by ptrace, etc.
	 * Get base of stack in r32.
	 */
	{
	 GET_THREAD_INFO(r32)
	 movei  r33, 0
	}

.Lretry_work_pending:
	/*
	 * Disable interrupts so as to make sure we don't
	 * miss an interrupt that sets any of the thread flags (like
	 * need_resched or sigpending) between sampling and the iret.
	 * Routines like schedule() or do_signal() may re-enable
	 * interrupts before returning.
	 */
	IRQ_DISABLE(r20, r21)
	TRACE_IRQS_OFF  /* Note: clobbers registers r0-r29 */


	/* Check to see if there is any work to do before returning to user. */
	{
	 addi   r29, r32, THREAD_INFO_FLAGS_OFFSET
	 moveli r1, lo16(_TIF_ALLWORK_MASK)
	}
	{
	 lw     r29, r29
	 auli   r1, r1, ha16(_TIF_ALLWORK_MASK)
	}
	and     r1, r29, r1
	bzt     r1, .Lrestore_all

	/*
	 * Make sure we have all the registers saved for signal
	 * handling, notify-resume, or single-step.  Call out to C
	 * code to figure out exactly what we need to do for each flag bit,
	 * then if necessary, reload the flags and recheck.
	 */
	{
	 PTREGS_PTR(r0, PTREGS_OFFSET_BASE)
	 bnz    r33, 1f
	}
	push_extra_callee_saves r0
	movei   r33, 1
1:	jal     do_work_pending
	bnz     r0, .Lretry_work_pending

	/*
	 * In the NMI case we
	 * omit the call to single_process_check_nohz, which normally checks
	 * to see if we should start or stop the scheduler tick, because
	 * we can't call arbitrary Linux code from an NMI context.
	 * We always call the homecache TLB deferral code to re-trigger
	 * the deferral mechanism.
	 *
	 * The other chunk of responsibility this code has is to reset the
	 * interrupt masks appropriately to reset irqs and NMIs.  We have
	 * to call TRACE_IRQS_OFF and TRACE_IRQS_ON to support all the
	 * lockdep-type stuff, but we can't set ICS until afterwards, since
	 * ICS can only be used in very tight chunks of code to avoid
	 * tripping over various assertions that it is off.
	 *
	 * (There is what looks like a window of vulnerability here since
	 * we might take a profile interrupt between the two SPR writes
	 * that set the mask, but since we write the low SPR word first,
	 * and our interrupt entry code checks the low SPR word, any
	 * profile interrupt will actually disable interrupts in both SPRs
	 * before returning, which is OK.)
	 */
.Lrestore_all:
	PTREGS_PTR(r0, PTREGS_OFFSET_EX1)
	{
	 lw     r0, r0
	 PTREGS_PTR(r32, PTREGS_OFFSET_FLAGS)
	}
	{
	 andi   r0, r0, SPR_EX_CONTEXT_1_1__PL_MASK
	 lw     r32, r32
	}
	bnz    r0, 1f
	j       2f
#if PT_FLAGS_DISABLE_IRQ != 1
# error Assuming PT_FLAGS_DISABLE_IRQ == 1 so we can use bbnst below
#endif
1:	bbnst   r32, 2f
	IRQ_DISABLE(r20,r21)
	TRACE_IRQS_OFF
	movei   r0, 1
	mtspr   INTERRUPT_CRITICAL_SECTION, r0
	bzt     r30, .Lrestore_regs
	j       3f
2:	TRACE_IRQS_ON
	movei   r0, 1
	mtspr   INTERRUPT_CRITICAL_SECTION, r0
	IRQ_ENABLE(r20, r21)
	bzt     r30, .Lrestore_regs
3:


	/*
	 * We now commit to returning from this interrupt, since we will be
	 * doing things like setting EX_CONTEXT SPRs and unwinding the stack
	 * frame.  No calls should be made to any other code after this point.
	 * This code should only be entered with ICS set.
	 * r32 must still be set to ptregs.flags.
	 * We launch loads to each cache line separately first, so we can
	 * get some parallelism out of the memory subsystem.
	 * We start zeroing caller-saved registers throughout, since
	 * that will save some cycles if this turns out to be a syscall.
	 */
.Lrestore_regs:
	FEEDBACK_REENTER(interrupt_return)   /* called from elsewhere */

	/*
	 * Rotate so we have one high bit and one low bit to test.
	 * - low bit says whether to restore all the callee-saved registers,
	 *   or just r30-r33, and r52 up.
	 * - high bit (i.e. sign bit) says whether to restore all the
	 *   caller-saved registers, or just r0.
	 */
#if PT_FLAGS_CALLER_SAVES != 2 || PT_FLAGS_RESTORE_REGS != 4
# error Rotate trick does not work :-)
#endif
	{
	 rli    r20, r32, 30
	 PTREGS_PTR(sp, PTREGS_OFFSET_REG(0))
	}

	/*
	 * Load cache lines 0, 2, and 3 in that order, then use
	 * the last loaded value, which makes it likely that the other
	 * cache lines have also loaded, at which point we should be
	 * able to safely read all the remaining words on those cache
	 * lines without waiting for the memory subsystem.
	 */
	pop_reg_zero r0, r28, sp, PTREGS_OFFSET_REG(30) - PTREGS_OFFSET_REG(0)
	pop_reg_zero r30, r2, sp, PTREGS_OFFSET_PC - PTREGS_OFFSET_REG(30)
	pop_reg_zero r21, r3, sp, PTREGS_OFFSET_EX1 - PTREGS_OFFSET_PC
	pop_reg_zero lr, r4, sp, PTREGS_OFFSET_REG(52) - PTREGS_OFFSET_EX1
	{
	 mtspr  SPR_EX_CONTEXT_K_0, r21
	 move   r5, zero
	}
	{
	 mtspr  SPR_EX_CONTEXT_K_1, lr
	 andi   lr, lr, SPR_EX_CONTEXT_1_1__PL_MASK  /* mask off ICS */
	}

	/* Restore callee-saveds that we actually use. */
	pop_reg_zero r52, r6, sp, PTREGS_OFFSET_REG(31) - PTREGS_OFFSET_REG(52)
	pop_reg_zero r31, r7
	pop_reg_zero r32, r8
	pop_reg_zero r33, r9, sp, PTREGS_OFFSET_REG(29) - PTREGS_OFFSET_REG(33)

	/*
	 * If we modified other callee-saveds, restore them now.
	 * This is rare, but could be via ptrace or signal handler.
	 */
	{
	 move   r10, zero
	 bbs    r20, .Lrestore_callees
	}
.Lcontinue_restore_regs:

	/* Check if we're returning from a syscall. */
	{
	 move   r11, zero
	 blzt   r20, 1f  /* no, so go restore callee-save registers */
	}

	/*
	 * Check if we're returning to userspace.
	 * Note that if we're not, we don't worry about zeroing everything.
	 */
	{
	 addli  sp, sp, PTREGS_OFFSET_LR - PTREGS_OFFSET_REG(29)
	 bnz    lr, .Lkernel_return
	}

	/*
	 * On return from syscall, we've restored r0 from pt_regs, but we
	 * clear the remainder of the caller-saved registers.  We could
	 * restore the syscall arguments, but there's not much point,
	 * and it ensures user programs aren't trying to use the
	 * caller-saves if we clear them, as well as avoiding leaking
	 * kernel pointers into userspace.
	 */
	pop_reg_zero lr, r12, sp, PTREGS_OFFSET_TP - PTREGS_OFFSET_LR
	pop_reg_zero tp, r13, sp, PTREGS_OFFSET_SP - PTREGS_OFFSET_TP
	{
	 lw     sp, sp
	 move   r14, zero
	 move   r15, zero
	}
	{ move r16, zero; move r17, zero }
	{ move r18, zero; move r19, zero }
	{ move r20, zero; move r21, zero }
	{ move r22, zero; move r23, zero }
	{ move r24, zero; move r25, zero }
	{ move r26, zero; move r27, zero }

	/* Set r1 to errno if we are returning an error, otherwise zero. */
	{
	 moveli r29, 4096
	 sub    r1, zero, r0
	}
	slt_u   r29, r1, r29
	{
	 mnz    r1, r29, r1
	 move   r29, zero
	}
	iret

	/*
	 * Not a syscall, so restore caller-saved registers.
	 * First kick off a load for cache line 1, which we're touching
	 * for the first time here.
	 */
	.align 64
1:	pop_reg r29, sp, PTREGS_OFFSET_REG(1) - PTREGS_OFFSET_REG(29)
	pop_reg r1
	pop_reg r2
	pop_reg r3
	pop_reg r4
	pop_reg r5
	pop_reg r6
	pop_reg r7
	pop_reg r8
	pop_reg r9
	pop_reg r10
	pop_reg r11
	pop_reg r12
	pop_reg r13
	pop_reg r14
	pop_reg r15
	pop_reg r16
	pop_reg r17
	pop_reg r18
	pop_reg r19
	pop_reg r20
	pop_reg r21
	pop_reg r22
	pop_reg r23
	pop_reg r24
	pop_reg r25
	pop_reg r26
	pop_reg r27
	pop_reg r28, sp, PTREGS_OFFSET_LR - PTREGS_OFFSET_REG(28)
	/* r29 already restored above */
	bnz     lr, .Lkernel_return
	pop_reg lr, sp, PTREGS_OFFSET_TP - PTREGS_OFFSET_LR
	pop_reg tp, sp, PTREGS_OFFSET_SP - PTREGS_OFFSET_TP
	lw      sp, sp
	iret

	/*
	 * We can't restore tp when in kernel mode, since a thread might
	 * have migrated from another cpu and brought a stale tp value.
	 */
.Lkernel_return:
	pop_reg lr, sp, PTREGS_OFFSET_SP - PTREGS_OFFSET_LR
	lw      sp, sp
	iret

	/* Restore callee-saved registers from r34 to r51. */
.Lrestore_callees:
	addli  sp, sp, PTREGS_OFFSET_REG(34) - PTREGS_OFFSET_REG(29)
	pop_reg r34
	pop_reg r35
	pop_reg r36
	pop_reg r37
	pop_reg r38
	pop_reg r39
	pop_reg r40
	pop_reg r41
	pop_reg r42
	pop_reg r43
	pop_reg r44
	pop_reg r45
	pop_reg r46
	pop_reg r47
	pop_reg r48
	pop_reg r49
	pop_reg r50
	pop_reg r51, sp, PTREGS_OFFSET_REG(29) - PTREGS_OFFSET_REG(51)
	j .Lcontinue_restore_regs
	STD_ENDPROC(interrupt_return)

	/*
	 * Some interrupts don't check for single stepping
	 */
	.pushsection .text.handle_interrupt_no_single_step,"ax"
handle_interrupt_no_single_step:
	finish_interrupt_save handle_interrupt_no_single_step
	{
	 jalr   r0
	 PTREGS_PTR(r0, PTREGS_OFFSET_BASE)
	}
	FEEDBACK_REENTER(handle_interrupt_no_single_step)
	{
	 movei  r30, 0   /* not an NMI */
	 j      interrupt_return
	}
	STD_ENDPROC(handle_interrupt_no_single_step)

	/*
	 * "NMI" interrupts mask ALL interrupts before calling the
	 * handler, and don't check thread flags, etc., on the way
	 * back out.  In general, the only things we do here for NMIs
	 * are the register save/restore, fixing the PC if we were
	 * doing single step, and the dataplane kernel-TLB management.
	 * We don't (for example) deal with start/stop of the sched tick.
	 */
	.pushsection .text.handle_nmi,"ax"
handle_nmi:
	finish_interrupt_save handle_nmi
	check_single_stepping normal, .Ldispatch_nmi
.Ldispatch_nmi:
	{
	 jalr   r0
	 PTREGS_PTR(r0, PTREGS_OFFSET_BASE)
	}
	FEEDBACK_REENTER(handle_nmi)
	j       interrupt_return
	STD_ENDPROC(handle_nmi)

	/*
	 * Parallel code for syscalls to handle_interrupt.
	 */
	.pushsection .text.handle_syscall,"ax"
handle_syscall:
	finish_interrupt_save handle_syscall

	/*
	 * Check for if we are single stepping in user level. If so, then
	 * we need to restore the PC.
	 */
	check_single_stepping syscall, .Ldispatch_syscall
.Ldispatch_syscall:

	/* Enable irqs. */
	TRACE_IRQS_ON
	IRQ_ENABLE(r20, r21)

	/* Bump the counter for syscalls made on this tile. */
	moveli  r20, lo16(irq_stat + IRQ_CPUSTAT_SYSCALL_COUNT_OFFSET)
	auli    r20, r20, ha16(irq_stat + IRQ_CPUSTAT_SYSCALL_COUNT_OFFSET)
	add     r20, r20, tp
	lw      r21, r20
	addi    r21, r21, 1
	{
	 sw     r20, r21
	 GET_THREAD_INFO(r31)
	}

	/* Trace syscalls, if requested. */
	addi	r31, r31, THREAD_INFO_FLAGS_OFFSET
	lw	r30, r31
	andi    r30, r30, _TIF_SYSCALL_TRACE
	bzt	r30, .Lrestore_syscall_regs
	{
	 PTREGS_PTR(r0, PTREGS_OFFSET_BASE)
	 jal    do_syscall_trace_enter
	}
	FEEDBACK_REENTER(handle_syscall)

	/*
	 * We always reload our registers from the stack at this
	 * point.  They might be valid, if we didn't build with
	 * TRACE_IRQFLAGS, and this isn't a dataplane tile, and we're not
	 * doing syscall tracing, but there are enough cases now that it
	 * seems simplest just to do the reload unconditionally.
	 */
.Lrestore_syscall_regs:
	PTREGS_PTR(r11, PTREGS_OFFSET_REG(0))
	pop_reg r0, r11
	pop_reg r1, r11
	pop_reg r2, r11
	pop_reg r3, r11
	pop_reg r4, r11
	pop_reg r5, r11, PTREGS_OFFSET_SYSCALL - PTREGS_OFFSET_REG(5)
	pop_reg TREG_SYSCALL_NR_NAME, r11

	/* Ensure that the syscall number is within the legal range. */
	moveli  r21, __NR_syscalls
	{
	 slt_u  r21, TREG_SYSCALL_NR_NAME, r21
	 moveli r20, lo16(sys_call_table)
	}
	{
	 bbns   r21, .Linvalid_syscall
	 auli   r20, r20, ha16(sys_call_table)
	}
	s2a     r20, TREG_SYSCALL_NR_NAME, r20
	lw      r20, r20

	/* Jump to syscall handler. */
	jalr    r20
.Lhandle_syscall_link: /* value of "lr" after "jalr r20" above */

	/*
	 * Write our r0 onto the stack so it gets restored instead
	 * of whatever the user had there before.
	 */
	PTREGS_PTR(r29, PTREGS_OFFSET_REG(0))
	sw      r29, r0

.Lsyscall_sigreturn_skip:
	FEEDBACK_REENTER(handle_syscall)

	/* Do syscall trace again, if requested. */
	lw	r30, r31
	andi    r30, r30, _TIF_SYSCALL_TRACE
	bzt     r30, 1f
	{
	 PTREGS_PTR(r0, PTREGS_OFFSET_BASE)
	 jal    do_syscall_trace_exit
	}
	FEEDBACK_REENTER(handle_syscall)
1:	{
	 movei  r30, 0               /* not an NMI */
	 j      .Lresume_userspace   /* jump into middle of interrupt_return */
	}

.Linvalid_syscall:
	/* Report an invalid syscall back to the user program */
	{
	 PTREGS_PTR(r29, PTREGS_OFFSET_REG(0))
	 movei  r28, -ENOSYS
	}
	sw      r29, r28
	{
	 movei  r30, 0               /* not an NMI */
	 j      .Lresume_userspace   /* jump into middle of interrupt_return */
	}
	STD_ENDPROC(handle_syscall)

	/* Return the address for oprofile to suppress in backtraces. */
STD_ENTRY_SECTION(handle_syscall_link_address, .text.handle_syscall)
	lnk     r0
	{
	 addli  r0, r0, .Lhandle_syscall_link - .
	 jrp    lr
	}
	STD_ENDPROC(handle_syscall_link_address)

STD_ENTRY(ret_from_fork)
	jal     sim_notify_fork
	jal     schedule_tail
	FEEDBACK_REENTER(ret_from_fork)
	{
	 movei  r30, 0               /* not an NMI */
	 j      .Lresume_userspace   /* jump into middle of interrupt_return */
	}
	STD_ENDPROC(ret_from_fork)

STD_ENTRY(ret_from_kernel_thread)
	jal     sim_notify_fork
	jal     schedule_tail
	FEEDBACK_REENTER(ret_from_fork)
	{
	 move   r0, r31
	 jalr   r30
	}
	FEEDBACK_REENTER(ret_from_kernel_thread)
	{
	 movei  r30, 0               /* not an NMI */
	 j      .Lresume_userspace   /* jump into middle of interrupt_return */
	}
	STD_ENDPROC(ret_from_kernel_thread)

	/*
	 * Code for ill interrupt.
	 */
	.pushsection .text.handle_ill,"ax"
handle_ill:
	finish_interrupt_save handle_ill

	/*
	 * Check for if we are single stepping in user level. If so, then
	 * we need to restore the PC.
	 */
	check_single_stepping ill, .Ldispatch_normal_ill

	{
	 /* See if the PC is the 1st bundle in the buffer */
	 seq    r25, r27, r26

	 /* Point to the 2nd bundle in the buffer */
	 addi   r26, r26, 8
	}
	{
	 /* Point to the original pc */
	 addi   r24, r29, SINGLESTEP_STATE_ORIG_PC_OFFSET

	 /* Branch if the PC is the 1st bundle in the buffer */
	 bnz    r25, 3f
	}
	{
	 /* See if the PC is the 2nd bundle of the buffer */
	 seq    r25, r27, r26

	 /* Set PC to next instruction */
	 addi   r24, r29, SINGLESTEP_STATE_NEXT_PC_OFFSET
	}
	{
	 /* Point to flags */
	 addi   r25, r29, SINGLESTEP_STATE_FLAGS_OFFSET

	 /* Branch if PC is in the second bundle */
	 bz     r25, 2f
	}
	/* Load flags */
	lw      r25, r25
	{
	 /*
	  * Get the offset for the register to restore
	  * Note: the lower bound is 2, so we have implicit scaling by 4.
	  *  No multiplication of the register number by the size of a register
	  *  is needed.
	  */
	 mm     r27, r25, zero, SINGLESTEP_STATE_TARGET_LB, \
		SINGLESTEP_STATE_TARGET_UB

	 /* Mask Rewrite_LR */
	 andi   r25, r25, SINGLESTEP_STATE_MASK_UPDATE
	}
	{
	 addi   r29, r29, SINGLESTEP_STATE_UPDATE_VALUE_OFFSET

	 /* Don't rewrite temp register */
	 bz     r25, 3f
	}
	{
	 /* Get the temp value */
	 lw     r29, r29

	 /* Point to where the register is stored */
	 add    r27, r27, sp
	}

	/* Add in the C ABI save area size to the register offset */
	addi    r27, r27, C_ABI_SAVE_AREA_SIZE

	/* Restore the user's register with the temp value */
	sw      r27, r29
	j       3f

2:
	/* Must be in the third bundle */
	addi    r24, r29, SINGLESTEP_STATE_BRANCH_NEXT_PC_OFFSET

3:
	/* set PC and continue */
	lw      r26, r24
	{
	 sw     r28, r26
	 GET_THREAD_INFO(r0)
	}

	/*
	 * Clear TIF_SINGLESTEP to prevent recursion if we execute an ill.
	 * The normal non-arch flow redundantly clears TIF_SINGLESTEP, but we
	 * need to clear it here and can't really impose on all other arches.
	 * So what's another write between friends?
	 */

	addi    r1, r0, THREAD_INFO_FLAGS_OFFSET
	{
	 lw     r2, r1
	 addi   r0, r0, THREAD_INFO_TASK_OFFSET  /* currently a no-op */
	}
	andi    r2, r2, ~_TIF_SINGLESTEP
	sw      r1, r2

	/* Issue a sigtrap */
	{
	 lw     r0, r0          /* indirect thru thread_info to get task_info*/
	 addi   r1, sp, C_ABI_SAVE_AREA_SIZE  /* put ptregs pointer into r1 */
	}

	jal     send_sigtrap    /* issue a SIGTRAP */
	FEEDBACK_REENTER(handle_ill)
	{
	 movei  r30, 0               /* not an NMI */
	 j      .Lresume_userspace   /* jump into middle of interrupt_return */
	}

.Ldispatch_normal_ill:
	{
	 jalr   r0
	 PTREGS_PTR(r0, PTREGS_OFFSET_BASE)
	}
	FEEDBACK_REENTER(handle_ill)
	{
	 movei  r30, 0   /* not an NMI */
	 j      interrupt_return
	}
	STD_ENDPROC(handle_ill)

/* Various stub interrupt handlers and syscall handlers */

STD_ENTRY_LOCAL(_kernel_double_fault)
	mfspr   r1, SPR_EX_CONTEXT_K_0
	move    r2, lr
	move    r3, sp
	move    r4, r52
	addi    sp, sp, -C_ABI_SAVE_AREA_SIZE
	j       kernel_double_fault
	STD_ENDPROC(_kernel_double_fault)

STD_ENTRY_LOCAL(bad_intr)
	mfspr   r2, SPR_EX_CONTEXT_K_0
	panic   "Unhandled interrupt %#x: PC %#lx"
	STD_ENDPROC(bad_intr)

/*
 * Special-case sigreturn to not write r0 to the stack on return.
 * This is technically more efficient, but it also avoids difficulties
 * in the 64-bit OS when handling 32-bit compat code, since we must not
 * sign-extend r0 for the sigreturn return-value case.
 */
#define PTREGS_SYSCALL_SIGRETURN(x, reg)                \
	STD_ENTRY(_##x);                                \
	addli   lr, lr, .Lsyscall_sigreturn_skip - .Lhandle_syscall_link; \
	{                                               \
	 PTREGS_PTR(reg, PTREGS_OFFSET_BASE);           \
	 j      x                                       \
	};                                              \
	STD_ENDPROC(_##x)

PTREGS_SYSCALL_SIGRETURN(sys_rt_sigreturn, r0)

/* Save additional callee-saves to pt_regs and jump to standard function. */
STD_ENTRY(_sys_clone)
	push_extra_callee_saves r4
	j       sys_clone
	STD_ENDPROC(_sys_clone)

/*
 * This entrypoint is taken for the cmpxchg and atomic_update fast
 * swints.  We may wish to generalize it to other fast swints at some
 * point, but for now there are just two very similar ones, which
 * makes it faster.
 *
 * The fast swint code is designed to have a small footprint.  It does
 * not save or restore any GPRs, counting on the caller-save registers
 * to be available to it on entry.  It does not modify any callee-save
 * registers (including "lr").  It does not check what PL it is being
 * called at, so you'd better not call it other than at PL0.
 * The <atomic.h> wrapper assumes it only clobbers r20-r29, so if
 * it ever is necessary to use more registers, be aware.
 *
 * It does not use the stack, but since it might be re-interrupted by
 * a page fault which would assume the stack was valid, it does
 * save/restore the stack pointer and zero it out to make sure it gets reset.
 * Since we always keep interrupts disabled, the hypervisor won't
 * clobber our EX_CONTEXT_K_x registers, so we don't save/restore them
 * (other than to advance the PC on return).
 *
 * We have to manually validate the user vs kernel address range
 * (since at PL1 we can read/write both), and for performance reasons
 * we don't allow cmpxchg on the fc000000 memory region, since we only
 * validate that the user address is below PAGE_OFFSET.
 *
 * We place it in the __HEAD section to ensure it is relatively
 * near to the intvec_SWINT_1 code (reachable by a conditional branch).
 *
 * Our use of ATOMIC_LOCK_REG here must match do_page_fault_ics().
 *
 * As we do in lib/atomic_asm_32.S, we bypass a store if the value we
 * would store is the same as the value we just loaded.
 */
	__HEAD
	.align 64
	/* Align much later jump on the start of a cache line. */
#if !ATOMIC_LOCKS_FOUND_VIA_TABLE()
	nop
#if PAGE_SIZE >= 0x10000
	nop
#endif
#endif
ENTRY(sys_cmpxchg)

	/*
	 * Save "sp" and set it zero for any possible page fault.
	 *
	 * HACK: We want to both zero sp and check r0's alignment,
	 * so we do both at once. If "sp" becomes nonzero we
	 * know r0 is unaligned and branch to the error handler that
	 * restores sp, so this is OK.
	 *
	 * ICS is disabled right now so having a garbage but nonzero
	 * sp is OK, since we won't execute any faulting instructions
	 * when it is nonzero.
	 */
	{
	 move   r27, sp
	 andi	sp, r0, 3
	}

	/*
	 * Get the lock address in ATOMIC_LOCK_REG, and also validate that the
	 * address is less than PAGE_OFFSET, since that won't trap at PL1.
	 * We only use bits less than PAGE_SHIFT to avoid having to worry
	 * about aliasing among multiple mappings of the same physical page,
	 * and we ignore the low 3 bits so we have one lock that covers
	 * both a cmpxchg64() and a cmpxchg() on either its low or high word.
	 * NOTE: this must match __atomic_hashed_lock() in lib/atomic_32.c.
	 */

#if (PAGE_OFFSET & 0xffff) != 0
# error Code here assumes PAGE_OFFSET can be loaded with just hi16()
#endif

#if ATOMIC_LOCKS_FOUND_VIA_TABLE()
	{
	 /* Check for unaligned input. */
	 bnz    sp, .Lcmpxchg_badaddr
	 mm     r25, r0, zero, 3, PAGE_SHIFT-1
	}
	{
	 crc32_32 r25, zero, r25
	 moveli r21, lo16(atomic_lock_ptr)
	}
	{
	 auli   r21, r21, ha16(atomic_lock_ptr)
	 auli   r23, zero, hi16(PAGE_OFFSET)  /* hugepage-aligned */
	}
	{
	 shri	r20, r25, 32 - ATOMIC_HASH_L1_SHIFT
	 slt_u  r23, r0, r23
	 lw	r26, r0  /* see comment in the "#else" for the "lw r26". */
	}
	{
	 s2a    r21, r20, r21
	 bbns   r23, .Lcmpxchg_badaddr
	}
	{
	 lw     r21, r21
	 seqi	r23, TREG_SYSCALL_NR_NAME, __NR_FAST_cmpxchg64
	 andi	r25, r25, ATOMIC_HASH_L2_SIZE - 1
	}
	{
	 /* Branch away at this point if we're doing a 64-bit cmpxchg. */
	 bbs    r23, .Lcmpxchg64
	 andi   r23, r0, 7       /* Precompute alignment for cmpxchg64. */
	}
	{
	 s2a	ATOMIC_LOCK_REG_NAME, r25, r21
	 j      .Lcmpxchg32_tns   /* see comment in the #else for the jump. */
	}

#else /* ATOMIC_LOCKS_FOUND_VIA_TABLE() */
	{
	 /* Check for unaligned input. */
	 bnz    sp, .Lcmpxchg_badaddr
	 auli   r23, zero, hi16(PAGE_OFFSET)  /* hugepage-aligned */
	}
	{
	 /*
	  * Slide bits into position for 'mm'. We want to ignore
	  * the low 3 bits of r0, and consider only the next
	  * ATOMIC_HASH_SHIFT bits.
	  * Because of C pointer arithmetic, we want to compute this:
	  *
	  * ((char*)atomic_locks +
	  *  (((r0 >> 3) & ((1 << ATOMIC_HASH_SHIFT) - 1)) << 2))
	  *
	  * Instead of two shifts we just ">> 1", and use 'mm'
	  * to ignore the low and high bits we don't want.
	  */
	 shri	r25, r0, 1

	 slt_u  r23, r0, r23

	 /*
	  * Ensure that the TLB is loaded before we take out the lock.
	  * On tilepro, this will start fetching the value all the way
	  * into our L1 as well (and if it gets modified before we
	  * grab the lock, it will be invalidated from our cache
	  * before we reload it).  On tile64, we'll start fetching it
	  * into our L1 if we're the home, and if we're not, we'll
	  * still at least start fetching it into the home's L2.
	  */
	 lw	r26, r0
	}
	{
	 auli	r21, zero, ha16(atomic_locks)

	 bbns   r23, .Lcmpxchg_badaddr
	}
#if PAGE_SIZE < 0x10000
	/* atomic_locks is page-aligned so for big pages we don't need this. */
	addli   r21, r21, lo16(atomic_locks)
#endif
	{
	 /*
	  * Insert the hash bits into the page-aligned pointer.
	  * ATOMIC_HASH_SHIFT is so big that we don't actually hash
	  * the unmasked address bits, as that may cause unnecessary
	  * collisions.
	  */
	 mm	ATOMIC_LOCK_REG_NAME, r25, r21, 2, (ATOMIC_HASH_SHIFT + 2) - 1

	 seqi	r23, TREG_SYSCALL_NR_NAME, __NR_FAST_cmpxchg64
	}
	{
	 /* Branch away at this point if we're doing a 64-bit cmpxchg. */
	 bbs    r23, .Lcmpxchg64
	 andi   r23, r0, 7       /* Precompute alignment for cmpxchg64. */
	}
	{
	 /*
	  * We very carefully align the code that actually runs with
	  * the lock held (twelve bundles) so that we know it is all in
	  * the icache when we start.  This instruction (the jump) is
	  * at the start of the first cache line, address zero mod 64;
	  * we jump to the very end of the second cache line to get that
	  * line loaded in the icache, then fall through to issue the tns
	  * in the third cache line, at which point it's all cached.
	  * Note that is for performance, not correctness.
	  */
	 j      .Lcmpxchg32_tns
	}

#endif /* ATOMIC_LOCKS_FOUND_VIA_TABLE() */

/* Symbol for do_page_fault_ics() to use to compare against the PC. */
.global __sys_cmpxchg_grab_lock
__sys_cmpxchg_grab_lock:

	/*
	 * Perform the actual cmpxchg or atomic_update.
	 */
.Ldo_cmpxchg32:
	{
	 lw     r21, r0
	 seqi	r23, TREG_SYSCALL_NR_NAME, __NR_FAST_atomic_update
	 move	r24, r2
	}
	{
	 seq    r22, r21, r1     /* See if cmpxchg matches. */
	 and	r25, r21, r1     /* If atomic_update, compute (*mem & mask) */
	}
	{
	 or	r22, r22, r23    /* Skip compare branch for atomic_update. */
	 add	r25, r25, r2     /* Compute (*mem & mask) + addend. */
	}
	{
	 mvnz	r24, r23, r25    /* Use atomic_update value if appropriate. */
	 bbns   r22, .Lcmpxchg32_nostore
	}
	seq     r22, r24, r21    /* Are we storing the value we loaded? */
	bbs     r22, .Lcmpxchg32_nostore
	sw      r0, r24

	/* The following instruction is the start of the second cache line. */
	/* Do slow mtspr here so the following "mf" waits less. */
	{
	 move   sp, r27
	 mtspr  SPR_EX_CONTEXT_K_0, r28
	}
	mf

	{
	 move   r0, r21
	 sw     ATOMIC_LOCK_REG_NAME, zero
	}
	iret

	/* Duplicated code here in the case where we don't overlap "mf" */
.Lcmpxchg32_nostore:
	{
	 move   r0, r21
	 sw     ATOMIC_LOCK_REG_NAME, zero
	}
	{
	 move   sp, r27
	 mtspr  SPR_EX_CONTEXT_K_0, r28
	}
	iret

	/*
	 * The locking code is the same for 32-bit cmpxchg/atomic_update,
	 * and for 64-bit cmpxchg.  We provide it as a macro and put
	 * it into both versions.  We can't share the code literally
	 * since it depends on having the right branch-back address.
	 */
	.macro  cmpxchg_lock, bitwidth

	/* Lock; if we succeed, jump back up to the read-modify-write. */
#ifdef CONFIG_SMP
	tns     r21, ATOMIC_LOCK_REG_NAME
#else
	/*
	 * Non-SMP preserves all the lock infrastructure, to keep the
	 * code simpler for the interesting (SMP) case.  However, we do
	 * one small optimization here and in atomic_asm.S, which is
	 * to fake out acquiring the actual lock in the atomic_lock table.
	 */
	movei	r21, 0
#endif

	/* Issue the slow SPR here while the tns result is in flight. */
	mfspr   r28, SPR_EX_CONTEXT_K_0

	{
	 addi   r28, r28, 8    /* return to the instruction after the swint1 */
	 bzt    r21, .Ldo_cmpxchg\bitwidth
	}
	/*
	 * The preceding instruction is the last thing that must be
	 * hot in the icache before we do the "tns" above.
	 */

#ifdef CONFIG_SMP
	/*
	 * We failed to acquire the tns lock on our first try.  Now use
	 * bounded exponential backoff to retry, like __atomic_spinlock().
	 */
	{
	 moveli r23, 2048       /* maximum backoff time in cycles */
	 moveli r25, 32         /* starting backoff time in cycles */
	}
1:	mfspr   r26, CYCLE_LOW  /* get start point for this backoff */
2:	mfspr   r22, CYCLE_LOW  /* test to see if we've backed off enough */
	sub     r22, r22, r26
	slt     r22, r22, r25
	bbst    r22, 2b
	{
	 shli   r25, r25, 1     /* double the backoff; retry the tns */
	 tns    r21, ATOMIC_LOCK_REG_NAME
	}
	slt     r26, r23, r25   /* is the proposed backoff too big? */
	{
	 mvnz   r25, r26, r23
	 bzt    r21, .Ldo_cmpxchg\bitwidth
	}
	j       1b
#endif /* CONFIG_SMP */
	.endm

.Lcmpxchg32_tns:
	/*
	 * This is the last instruction on the second cache line.
	 * The nop here loads the second line, then we fall through
	 * to the tns to load the third line before we take the lock.
	 */
	nop
	cmpxchg_lock 32

	/*
	 * This code is invoked from sys_cmpxchg after most of the
	 * preconditions have been checked.  We still need to check
	 * that r0 is 8-byte aligned, since if it's not we won't
	 * actually be atomic.  However, ATOMIC_LOCK_REG has the atomic
	 * lock pointer and r27/r28 have the saved SP/PC.
	 * r23 is holding "r0 & 7" so we can test for alignment.
	 * The compare value is in r2/r3; the new value is in r4/r5.
	 * On return, we must put the old value in r0/r1.
	 */
	.align 64
.Lcmpxchg64:
	{
#if ATOMIC_LOCKS_FOUND_VIA_TABLE()
	 s2a	ATOMIC_LOCK_REG_NAME, r25, r21
#endif
	 bzt     r23, .Lcmpxchg64_tns
	}
	j       .Lcmpxchg_badaddr

.Ldo_cmpxchg64:
	{
	 lw     r21, r0
	 addi   r25, r0, 4
	}
	{
	 lw     r1, r25
	}
	seq     r26, r21, r2
	{
	 bz     r26, .Lcmpxchg64_mismatch
	 seq    r26, r1, r3
	}
	{
	 bz     r26, .Lcmpxchg64_mismatch
	}
	sw      r0, r4
	sw      r25, r5

	/*
	 * The 32-bit path provides optimized "match" and "mismatch"
	 * iret paths, but we don't have enough bundles in this cache line
	 * to do that, so we just make even the "mismatch" path do an "mf".
	 */
.Lcmpxchg64_mismatch:
	{
	 move   sp, r27
	 mtspr  SPR_EX_CONTEXT_K_0, r28
	}
	mf
	{
	 move   r0, r21
	 sw     ATOMIC_LOCK_REG_NAME, zero
	}
	iret

.Lcmpxchg64_tns:
	cmpxchg_lock 64


	/*
	 * Reset sp and revector to sys_cmpxchg_badaddr(), which will
	 * just raise the appropriate signal and exit.  Doing it this
	 * way means we don't have to duplicate the code in intvec.S's
	 * int_hand macro that locates the top of the stack.
	 */
.Lcmpxchg_badaddr:
	{
	 moveli TREG_SYSCALL_NR_NAME, __NR_cmpxchg_badaddr
	 move   sp, r27
	}
	j       intvec_SWINT_1
	ENDPROC(sys_cmpxchg)
	ENTRY(__sys_cmpxchg_end)


/* The single-step support may need to read all the registers. */
int_unalign:
	push_extra_callee_saves r0
	j       do_trap

/* Include .intrpt1 array of interrupt vectors */
	.section ".intrpt1", "ax"

#define op_handle_perf_interrupt bad_intr
#define op_handle_aux_perf_interrupt bad_intr

#ifndef CONFIG_HARDWALL
#define do_hardwall_trap bad_intr
#endif

	int_hand     INT_ITLB_MISS, ITLB_MISS, \
		     do_page_fault, handle_interrupt_no_single_step
	int_hand     INT_MEM_ERROR, MEM_ERROR, bad_intr
	int_hand     INT_ILL, ILL, do_trap, handle_ill
	int_hand     INT_GPV, GPV, do_trap
	int_hand     INT_SN_ACCESS, SN_ACCESS, do_trap
	int_hand     INT_IDN_ACCESS, IDN_ACCESS, do_trap
	int_hand     INT_UDN_ACCESS, UDN_ACCESS, do_trap
	int_hand     INT_IDN_REFILL, IDN_REFILL, bad_intr
	int_hand     INT_UDN_REFILL, UDN_REFILL, bad_intr
	int_hand     INT_IDN_COMPLETE, IDN_COMPLETE, bad_intr
	int_hand     INT_UDN_COMPLETE, UDN_COMPLETE, bad_intr
	int_hand     INT_SWINT_3, SWINT_3, do_trap
	int_hand     INT_SWINT_2, SWINT_2, do_trap
	int_hand     INT_SWINT_1, SWINT_1, SYSCALL, handle_syscall
	int_hand     INT_SWINT_0, SWINT_0, do_trap
	int_hand     INT_UNALIGN_DATA, UNALIGN_DATA, int_unalign
	int_hand     INT_DTLB_MISS, DTLB_MISS, do_page_fault
	int_hand     INT_DTLB_ACCESS, DTLB_ACCESS, do_page_fault
	int_hand     INT_DMATLB_MISS, DMATLB_MISS, do_page_fault
	int_hand     INT_DMATLB_ACCESS, DMATLB_ACCESS, do_page_fault
	int_hand     INT_SNITLB_MISS, SNITLB_MISS, do_page_fault
	int_hand     INT_SN_NOTIFY, SN_NOTIFY, bad_intr
	int_hand     INT_SN_FIREWALL, SN_FIREWALL, do_hardwall_trap
	int_hand     INT_IDN_FIREWALL, IDN_FIREWALL, bad_intr
	int_hand     INT_UDN_FIREWALL, UDN_FIREWALL, do_hardwall_trap
	int_hand     INT_TILE_TIMER, TILE_TIMER, do_timer_interrupt
	int_hand     INT_IDN_TIMER, IDN_TIMER, bad_intr
	int_hand     INT_UDN_TIMER, UDN_TIMER, bad_intr
	int_hand     INT_DMA_NOTIFY, DMA_NOTIFY, bad_intr
	int_hand     INT_IDN_CA, IDN_CA, bad_intr
	int_hand     INT_UDN_CA, UDN_CA, bad_intr
	int_hand     INT_IDN_AVAIL, IDN_AVAIL, bad_intr
	int_hand     INT_UDN_AVAIL, UDN_AVAIL, bad_intr
	int_hand     INT_PERF_COUNT, PERF_COUNT, \
		     op_handle_perf_interrupt, handle_nmi
	int_hand     INT_INTCTRL_3, INTCTRL_3, bad_intr
#if CONFIG_KERNEL_PL == 2
	dc_dispatch  INT_INTCTRL_2, INTCTRL_2
	int_hand     INT_INTCTRL_1, INTCTRL_1, bad_intr
#else
	int_hand     INT_INTCTRL_2, INTCTRL_2, bad_intr
	dc_dispatch  INT_INTCTRL_1, INTCTRL_1
#endif
	int_hand     INT_INTCTRL_0, INTCTRL_0, bad_intr
	int_hand     INT_MESSAGE_RCV_DWNCL, MESSAGE_RCV_DWNCL, \
		     hv_message_intr
	int_hand     INT_DEV_INTR_DWNCL, DEV_INTR_DWNCL, \
		     tile_dev_intr
	int_hand     INT_I_ASID, I_ASID, bad_intr
	int_hand     INT_D_ASID, D_ASID, bad_intr
	int_hand     INT_DMATLB_MISS_DWNCL, DMATLB_MISS_DWNCL, \
		     do_page_fault
	int_hand     INT_SNITLB_MISS_DWNCL, SNITLB_MISS_DWNCL, \
		     do_page_fault
	int_hand     INT_DMATLB_ACCESS_DWNCL, DMATLB_ACCESS_DWNCL, \
		     do_page_fault
	int_hand     INT_SN_CPL, SN_CPL, bad_intr
	int_hand     INT_DOUBLE_FAULT, DOUBLE_FAULT, do_trap
#if CHIP_HAS_AUX_PERF_COUNTERS()
	int_hand     INT_AUX_PERF_COUNT, AUX_PERF_COUNT, \
		     op_handle_aux_perf_interrupt, handle_nmi
#endif

	/* Synthetic interrupt delivered only by the simulator */
	int_hand     INT_BREAKPOINT, BREAKPOINT, do_breakpoint