summaryrefslogtreecommitdiffstats
path: root/arch/tile/kernel/intvec_64.S
blob: 800b91d3f9dc79f15b22a08e3bba21f89640268a (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
/*
 * Copyright 2011 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/unistd.h>
#include <linux/init.h>
#include <asm/ptrace.h>
#include <asm/thread_info.h>
#include <asm/irqflags.h>
#include <asm/asm-offsets.h>
#include <asm/types.h>
#include <asm/traps.h>
#include <asm/signal.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 CONFIG_KERNEL_PL == 1 || CONFIG_KERNEL_PL == 2
/*
 * Set "result" non-zero if ex1 holds the PL of the kernel
 * (with or without ICS being set).  Note this works only
 * because we never find the PL at level 3.
 */
# define IS_KERNEL_EX1(result, ex1) andi result, ex1, CONFIG_KERNEL_PL
#else
# error Recode IS_KERNEL_EX1 for CONFIG_KERNEL_PL
#endif

	.macro  push_reg reg, ptr=sp, delta=-8
	{
	 st     \ptr, \reg
	 addli  \ptr, \ptr, \delta
	}
	.endm

	.macro  pop_reg reg, ptr=sp, delta=8
	{
	 ld     \reg, \ptr
	 addli  \ptr, \ptr, \delta
	}
	.endm

	.macro  pop_reg_zero reg, zreg, ptr=sp, delta=8
	{
	 move   \zreg, zero
	 ld     \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, hw2_last(1b)
	}
	{
	 shl16insli r0, r0, hw1(1b)
	}
	{
	 shl16insli r0, r0, hw0(1b)
	 jal    panic
	}
	.endm

	/*
	 * Unalign data exception fast handling: In order to handle
	 * unaligned data access, a fast JIT version is generated and stored
	 * in a specific area in user space. We first need to do a quick poke
	 * to see if the JIT is available. We use certain bits in the fault
	 * PC (3 to 9 is used for 16KB page size) as index to address the JIT
	 * code area. The first 64bit word is the fault PC, and the 2nd one is
	 * the fault bundle itself. If these 2 words both match, then we
	 * directly "iret" to JIT code. If not, a slow path is invoked to
	 * generate new JIT code. Note: the current JIT code WILL be
	 * overwritten if it existed. So, ideally we can handle 128 unalign
	 * fixups via JIT. For lookup efficiency and to effectively support
	 * tight loops with multiple unaligned reference, a simple
	 * direct-mapped cache is used.
	 *
	 * SPR_EX_CONTEXT_K_0 is modified to return to JIT code.
	 * SPR_EX_CONTEXT_K_1 has ICS set.
	 * SPR_EX_CONTEXT_0_0 is setup to user program's next PC.
	 * SPR_EX_CONTEXT_0_1 = 0.
	 */
	.macro int_hand_unalign_fast  vecnum, vecname
	.org  (\vecnum << 8)
intvec_\vecname:
	/* Put r3 in SPR_SYSTEM_SAVE_K_1.  */
	mtspr   SPR_SYSTEM_SAVE_K_1, r3

	mfspr   r3, SPR_EX_CONTEXT_K_1
	/*
	 * Examine if exception comes from user without ICS set.
	 * If not, just go directly to the slow path.
	 */
	bnez    r3, hand_unalign_slow_nonuser

	mfspr   r3, SPR_SYSTEM_SAVE_K_0

	/* Get &thread_info->unalign_jit_tmp[0] in r3. */
	bfexts  r3, r3, 0, CPU_SHIFT-1
	mm      r3, zero, LOG2_THREAD_SIZE, 63
	addli   r3, r3, THREAD_INFO_UNALIGN_JIT_TMP_OFFSET

	/*
	 * Save r0, r1, r2 into thread_info array r3 points to
	 * from low to high memory in order.
	 */
	st_add  r3, r0, 8
	st_add  r3, r1, 8
	{
	 st_add r3, r2, 8
	 andi   r2, sp, 7
	}

	/* Save stored r3 value so we can revert it on a page fault. */
	mfspr   r1, SPR_SYSTEM_SAVE_K_1
	st      r3, r1

	{
	 /* Generate a SIGBUS if sp is not 8-byte aligned. */
	 bnez   r2, hand_unalign_slow_badsp
	}

	/*
	 * Get the thread_info in r0; load r1 with pc. Set the low bit of sp
	 * as an indicator to the page fault code in case we fault.
	 */
	{
	 ori    sp, sp, 1
	 mfspr  r1, SPR_EX_CONTEXT_K_0
	}

	/* Add the jit_info offset in thread_info; extract r1 [3:9] into r2. */
	{
	 addli  r0, r3, THREAD_INFO_UNALIGN_JIT_BASE_OFFSET - \
	  (THREAD_INFO_UNALIGN_JIT_TMP_OFFSET + (3 * 8))
	 bfextu r2, r1, 3, (2 + PAGE_SHIFT - UNALIGN_JIT_SHIFT)
	}

	/* Load the jit_info; multiply r2 by 128. */
	{
	 ld     r0, r0
	 shli   r2, r2, UNALIGN_JIT_SHIFT
	}

	/*
	 * If r0 is NULL, the JIT page is not mapped, so go to slow path;
	 * add offset r2 to r0 at the same time.
	 */
	{
	 beqz   r0, hand_unalign_slow
	 add    r2, r0, r2
	}

        /*
	 * We are loading from userspace (both the JIT info PC and
	 * instruction word, and the instruction word we executed)
	 * and since either could fault while holding the interrupt
	 * critical section, we must tag this region and check it in
	 * do_page_fault() to handle it properly.
	 */
ENTRY(__start_unalign_asm_code)

	/* Load first word of JIT in r0 and increment r2 by 8. */
	ld_add  r0, r2, 8

	/*
	 * Compare the PC with the 1st word in JIT; load the fault bundle
	 * into r1.
	 */
	{
	 cmpeq  r0, r0, r1
	 ld     r1, r1
	}

	/* Go to slow path if PC doesn't match. */
	beqz    r0, hand_unalign_slow

	/*
	 * Load the 2nd word of JIT, which is supposed to be the fault
	 * bundle for a cache hit. Increment r2; after this bundle r2 will
	 * point to the potential start of the JIT code we want to run.
	 */
	ld_add  r0, r2, 8

	/* No further accesses to userspace are done after this point. */
ENTRY(__end_unalign_asm_code)

	/* Compare the real bundle with what is saved in the JIT area. */
	{
	 cmpeq  r0, r1, r0
	 mtspr  SPR_EX_CONTEXT_0_1, zero
	}

	/* Go to slow path if the fault bundle does not match. */
	beqz    r0, hand_unalign_slow

	/*
	 * A cache hit is found.
	 * r2 points to start of JIT code (3rd word).
	 * r0 is the fault pc.
	 * r1 is the fault bundle.
	 * Reset the low bit of sp.
	 */
	{
	 mfspr  r0, SPR_EX_CONTEXT_K_0
	 andi   sp, sp, ~1
	}

	/* Write r2 into EX_CONTEXT_K_0 and increment PC. */
	{
	 mtspr  SPR_EX_CONTEXT_K_0, r2
	 addi   r0, r0, 8
	}

	/*
	 * Set ICS on kernel EX_CONTEXT_K_1 in order to "iret" to
	 * user with ICS set. This way, if the JIT fixup causes another
	 * unalign exception (which shouldn't be possible) the user
	 * process will be terminated with SIGBUS. Also, our fixup will
	 * run without interleaving with external interrupts.
	 * Each fixup is at most 14 bundles, so it won't hold ICS for long.
	 */
	{
	 movei  r1, PL_ICS_EX1(USER_PL, 1)
	 mtspr  SPR_EX_CONTEXT_0_0, r0
	}

	{
	 mtspr  SPR_EX_CONTEXT_K_1, r1
	 addi   r3, r3, -(3 * 8)
	}

	/* Restore r0..r3. */
	ld_add  r0, r3, 8
	ld_add  r1, r3, 8
	ld_add  r2, r3, 8
	ld      r3, r3

	iret
	ENDPROC(intvec_\vecname)
	.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
intvec_\vecname:
	/* 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 unalign data fastpath code sets the low bit in sp to
	 * force us to reset it here on fault.
	 */
	{
	 blbs   sp, 2f
	 IS_KERNEL_EX1(r0, r0)
	}

	.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.
	 */
	beqz    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.
	 */
	{
	 bnez   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.
	 *
	 * FIXME: if we end up never using this path, perhaps we should
	 * prevent the hypervisor from generating downcalls in this case.
	 * The advantage of getting a downcall is we can panic in Linux.
	 */
	mfspr   r0, SPR_SYSTEM_SAVE_K_2
	{
	 bltz   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 high bits, and
	 * the current stack top in the lower bits.  So we recover
	 * our starting stack value by sign-extending the low bits, then
	 * point sp at the top aligned address on the actual stack page.
	 */
	mfspr   r0, SPR_SYSTEM_SAVE_K_0
	bfexts  r0, r0, 0, CPU_SHIFT-1

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 8 just below the r0 value this "andi" computes.
	 * Note that we never write to cache line 8, and we skip
	 * cache lines 1-3 for syscalls.
	 *
	 *    cache line 8: ptregs padding (two words)
	 *    cache line 7: sp, lr, pc, ex1, faultnum, orig_r0, flags, cmpexch
	 *    cache line 6: r46...r53 (tp)
	 *    cache line 5: r38...r45
	 *    cache line 4: r30...r37
	 *    cache line 3: r22...r29
	 *    cache line 2: r14...r21
	 *    cache line 1: r6...r13
	 *    cache line 0: 2 x frame, r0..r5
	 */
#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,
	 * and needs sp set to its final location at the bottom of
	 * the stack frame.
	 */
	addli   r0, r0, PTREGS_OFFSET_LR - (PTREGS_SIZE + KSTK_PTREGS_GAP)
	wh64    r0   /* cache line 7 */
	{
	 st     r0, lr
	 addli  r0, r0, PTREGS_OFFSET_SP - PTREGS_OFFSET_LR
	}
	{
	 st     r0, sp
	 addli  sp, r0, PTREGS_OFFSET_REG(52) - PTREGS_OFFSET_SP
	}
	wh64    sp   /* cache line 6 */
	{
	 st     sp, r52
	 addli  sp, sp, PTREGS_OFFSET_REG(1) - PTREGS_OFFSET_REG(52)
	}
	wh64    sp   /* cache line 0 */
	{
	 st     sp, r1
	 addli  sp, sp, PTREGS_OFFSET_REG(2) - PTREGS_OFFSET_REG(1)
	}
	{
	 st     sp, r2
	 addli  sp, sp, PTREGS_OFFSET_REG(3) - PTREGS_OFFSET_REG(2)
	}
	{
	 st     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
	{
	 st     sp, r0
	 addli  sp, sp, PTREGS_OFFSET_EX1 - PTREGS_OFFSET_PC
	}
	mfspr   r0, SPR_EX_CONTEXT_K_1
	{
	 st     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
	}
	{
	 st     sp, r0
	 .else
	 movei  r1, \vecnum
	}
	{
	 st     sp, r1
	 .endif
	 addli  sp, sp, PTREGS_OFFSET_REG(0) - PTREGS_OFFSET_FAULTNUM
	}
	mfspr   r0, SPR_SYSTEM_SAVE_K_1    /* Original r0 */
	{
	 st     sp, r0
	 addi   sp, sp, -PTREGS_OFFSET_REG(0) - 8
	}
	{
	 st     sp, zero        /* write zero into "Next SP" frame pointer */
	 addi   sp, sp, -8      /* leave SP pointing at bottom of frame */
	}
	.ifc \processing,handle_syscall
	j       handle_syscall
	.else
	/* Capture per-interrupt SPR context to registers. */
	.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_ILL_TRANS
	mfspr   r2, ILL_VA_PC
	.else
	.ifc \vecnum, INT_DOUBLE_FAULT
	mfspr   r2, SPR_SYSTEM_SAVE_K_2   /* double fault info from HV */
	.else
	.ifc \c_routine, do_trap
	mfspr   r2, GPV_REASON
	.else
	.ifc \c_routine, handle_perf_interrupt
	mfspr   r2, PERF_COUNT_STS
	.else
	.ifc \c_routine, handle_perf_interrupt
	mfspr   r2, AUX_PERF_COUNT_STS
	.endif
	.ifc \c_routine, do_nmi
	mfspr   r2, SPR_SYSTEM_SAVE_K_2   /* nmi type */
	.else
	.endif
	.endif
	.endif
	.endif
	.endif
	.endif
	/* Put function pointer in r0 */
	moveli  r0, hw2_last(\c_routine)
	shl16insli r0, r0, hw1(\c_routine)
	{
	 shl16insli r0, r0, hw0(\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, .intrpt, 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
	 st     r52, r0
	 .else
	 st     r52, zero
	 .endif
	 PTREGS_PTR(r52, PTREGS_OFFSET_TP)
	}
	st      r52, tp
	{
	 mfspr  tp, CMPEXCH_VALUE
	 PTREGS_PTR(r52, PTREGS_OFFSET_CMPEXCH)
	}

	/*
	 * 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.)
	 */
	{
	 st     r52, tp
	 PTREGS_PTR(r52, PTREGS_OFFSET_REG(33))
	}
	wh64    r52    /* cache line 4 */
	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 3 */
	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
	wh64    r52   /* cache line 2 */
	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
	wh64    r52   /* cache line 1 */
	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
	st      r52, r4

	/*
	 * 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 kernel pl with
	 * irqs disabled.
	 */
	mfspr   r32, SPR_EX_CONTEXT_K_1
	{
	 IS_KERNEL_EX1(r22, r22)
	 PTREGS_PTR(r21, PTREGS_OFFSET_FLAGS)
	}
	beqzt   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
	st      r21, r32

	/*
	 * 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.
	 *
	 * It's important that no code before this point touch memory
	 * other than our own stack (to keep the invariant that this
	 * is all that gets touched under ICS), and that no code after
	 * this point reference any interrupt-specific SPR, in particular
	 * the EX_CONTEXT_K_ values.
	 */
	.ifc \function,handle_nmi
	IRQ_DISABLE_ALL(r20)
	.else
	IRQ_DISABLE(r20, r21)
	.endif
	mtspr   INTERRUPT_CRITICAL_SECTION, zero

	/* Load tp with our per-cpu offset. */
#ifdef CONFIG_SMP
	{
	 mfspr  r20, SPR_SYSTEM_SAVE_K_0
	 moveli r21, hw2_last(__per_cpu_offset)
	}
	{
	 shl16insli r21, r21, hw1(__per_cpu_offset)
	 bfextu r20, r20, CPU_SHIFT, 63
	}
	shl16insli r21, r21, hw0(__per_cpu_offset)
	shl3add r20, r20, r21
	ld      tp, r20
#else
	move    tp, zero
#endif

#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
	moveli  r21, hw2_last(intvec_feedback)
	shl16insli r21, r21, hw1(intvec_feedback)
	shl16insli r21, r21, hw0(intvec_feedback)
	add     r20, r20, r21
	jalr    r20

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

	/*
	 * Prepare the first 256 stack bytes to be rapidly accessible
	 * without having to fetch the background data.
	 */
	addi    r52, sp, -64
	{
	 wh64   r52
	 addi   r52, r52, -64
	}
	{
	 wh64   r52
	 addi   r52, r52, -64
	}
	{
	 wh64   r52
	 addi   r52, r52, -64
	}
	wh64    r52

#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

	/*
	 * 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

	/* 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. */
	{
	 bnez   r30, .Lrestore_all  /* NMIs don't special-case user-space */
	 PTREGS_PTR(r29, PTREGS_OFFSET_EX1)
	}
	ld      r29, r29
	IS_KERNEL_EX1(r29, r29)
	{
	 beqzt  r29, .Lresume_userspace
	 move   r29, sp
	}

#ifdef CONFIG_PREEMPT
	/* Returning to kernel space. Check if we need preemption. */
	EXTRACT_THREAD_INFO(r29)
	addli   r28, r29, THREAD_INFO_FLAGS_OFFSET
	{
	 ld     r28, r28
	 addli  r29, r29, THREAD_INFO_PREEMPT_COUNT_OFFSET
	}
	{
	 andi   r28, r28, _TIF_NEED_RESCHED
	 ld4s   r29, r29
	}
	beqzt   r28, 1f
	bnez    r29, 1f
	/* Disable interrupts explicitly for preemption. */
	IRQ_DISABLE(r20,r21)
	TRACE_IRQS_OFF
	jal     preempt_schedule_irq
	FEEDBACK_REENTER(interrupt_return)
1:
#endif

	/* If we're resuming to _cpu_idle_nap, bump PC forward by 8. */
	{
	 moveli r27, hw2_last(_cpu_idle_nap)
	 PTREGS_PTR(r29, PTREGS_OFFSET_PC)
	}
	{
	 ld     r28, r29
	 shl16insli r27, r27, hw1(_cpu_idle_nap)
	}
	{
	 shl16insli r27, r27, hw0(_cpu_idle_nap)
	}
	{
	 cmpeq  r27, r27, r28
	}
	{
	 blbc   r27, .Lrestore_all
	 addi   r28, r28, 8
	}
	st      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.
	 */
	{
	 movei  r33, 0
	 move   r32, sp
	}

	/* Get base of stack in r32. */
	EXTRACT_THREAD_INFO(r32)

.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, hw1_last(_TIF_ALLWORK_MASK)
	}
	{
	 ld     r29, r29
	 shl16insli r1, r1, hw0(_TIF_ALLWORK_MASK)
	}
	and     r1, r29, r1
	beqzt   r1, .Lrestore_all

	/*
	 * Make sure we have all the registers saved for signal
	 * handling or notify-resume.  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)
	 bnez   r33, 1f
	}
	push_extra_callee_saves r0
	movei   r33, 1
1:	jal     do_work_pending
	bnez    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.
	 */
.Lrestore_all:
	PTREGS_PTR(r0, PTREGS_OFFSET_EX1)
	{
	 ld      r0, r0
	 PTREGS_PTR(r32, PTREGS_OFFSET_FLAGS)
	}
	{
	 IS_KERNEL_EX1(r0, r0)
	 ld     r32, r32
	}
	bnez    r0, 1f
	j       2f
#if PT_FLAGS_DISABLE_IRQ != 1
# error Assuming PT_FLAGS_DISABLE_IRQ == 1 so we can use blbct below
#endif
1:	blbct   r32, 2f
	IRQ_DISABLE(r20,r21)
	TRACE_IRQS_OFF
	movei   r0, 1
	mtspr   INTERRUPT_CRITICAL_SECTION, r0
	beqzt   r30, .Lrestore_regs
	j       3f
2:	TRACE_IRQS_ON
	IRQ_ENABLE_LOAD(r20, r21)
	movei   r0, 1
	mtspr   INTERRUPT_CRITICAL_SECTION, r0
	IRQ_ENABLE_APPLY(r20, r21)
	beqzt   r30, .Lrestore_regs
3:

#if INT_PERF_COUNT + 1 != INT_AUX_PERF_COUNT
# error Bad interrupt assumption
#endif
	{
	 movei  r0, 3   /* two adjacent bits for the PERF_COUNT mask */
	 beqz   r31, .Lrestore_regs
	}
	shli    r0, r0, INT_PERF_COUNT
	mtspr   SPR_INTERRUPT_MASK_RESET_K, r0

	/*
	 * 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:

	/*
	 * 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
	{
	 rotli  r20, r32, 62
	 PTREGS_PTR(sp, PTREGS_OFFSET_REG(0))
	}

	/*
	 * Load cache lines 0, 4, 6 and 7, 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 r0, sp, PTREGS_OFFSET_REG(30) - PTREGS_OFFSET_REG(0)
	pop_reg r30, sp, PTREGS_OFFSET_REG(52) - PTREGS_OFFSET_REG(30)
	pop_reg_zero r52, r3, sp, PTREGS_OFFSET_CMPEXCH - PTREGS_OFFSET_REG(52)
	pop_reg_zero r21, r27, sp, PTREGS_OFFSET_EX1 - PTREGS_OFFSET_CMPEXCH
	pop_reg_zero lr, r2, sp, PTREGS_OFFSET_PC - PTREGS_OFFSET_EX1
	{
	 mtspr  CMPEXCH_VALUE, r21
	 move   r4, zero
	}
	pop_reg r21, sp, PTREGS_OFFSET_REG(31) - PTREGS_OFFSET_PC
	{
	 mtspr  SPR_EX_CONTEXT_K_1, lr
	 IS_KERNEL_EX1(lr, lr)
	}
	{
	 mtspr  SPR_EX_CONTEXT_K_0, r21
	 move   r5, zero
	}

	/* Restore callee-saveds that we actually use. */
	pop_reg_zero r31, r6
	pop_reg_zero r32, r7
	pop_reg_zero r33, r8, 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   r9, zero
	 blbs   r20, .Lrestore_callees
	}
.Lcontinue_restore_regs:

	/* Check if we're returning from a syscall. */
	{
	 move   r10, zero
	 bltzt  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)
	 bnez   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, r11, sp, PTREGS_OFFSET_TP - PTREGS_OFFSET_LR
	pop_reg_zero tp, r12, sp, PTREGS_OFFSET_SP - PTREGS_OFFSET_TP
	{
	 ld     sp, sp
	 move   r13, zero
	 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 }

	/* Set r1 to errno if we are returning an error, otherwise zero. */
	{
	 moveli r29, 4096
	 sub    r1, zero, r0
	}
	{
	 move   r28, zero
	 cmpltu r29, r1, r29
	}
	{
	 mnz    r1, r29, r1
	 move   r29, zero
	}
	iret

	/*
	 * Not a syscall, so restore caller-saved registers.
	 * First kick off loads for cache lines 1-3, which we're touching
	 * for the first time here.
	 */
	.align 64
1:	pop_reg r29, sp, PTREGS_OFFSET_REG(21) - PTREGS_OFFSET_REG(29)
	pop_reg r21, sp, PTREGS_OFFSET_REG(13) - PTREGS_OFFSET_REG(21)
	pop_reg r13, sp, PTREGS_OFFSET_REG(1) - PTREGS_OFFSET_REG(13)
	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, sp, 16
	/* r13 already restored above */
	pop_reg r14
	pop_reg r15
	pop_reg r16
	pop_reg r17
	pop_reg r18
	pop_reg r19
	pop_reg r20, sp, 16
	/* r21 already restored above */
	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 */
	bnez    lr, .Lkernel_return
	pop_reg lr, sp, PTREGS_OFFSET_TP - PTREGS_OFFSET_LR
	pop_reg tp, sp, PTREGS_OFFSET_SP - PTREGS_OFFSET_TP
	ld      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
	ld      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)

	/*
	 * "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 register save/restore and 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
	{
	 jalr   r0
	 PTREGS_PTR(r0, PTREGS_OFFSET_BASE)
	}
	FEEDBACK_REENTER(handle_nmi)
	{
	 movei  r30, 1
	 cmpeq  r31, r0, zero
	}
	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

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

	/* Bump the counter for syscalls made on this tile. */
	moveli r20, hw2_last(irq_stat + IRQ_CPUSTAT_SYSCALL_COUNT_OFFSET)
	shl16insli r20, r20, hw1(irq_stat + IRQ_CPUSTAT_SYSCALL_COUNT_OFFSET)
	shl16insli r20, r20, hw0(irq_stat + IRQ_CPUSTAT_SYSCALL_COUNT_OFFSET)
	add     r20, r20, tp
	ld4s    r21, r20
	{
	 addi   r21, r21, 1
	 move   r31, sp
	}
	{
	 st4    r20, r21
	 EXTRACT_THREAD_INFO(r31)
	}

	/* Trace syscalls, if requested. */
	addi	r31, r31, THREAD_INFO_FLAGS_OFFSET
	{
	 ld     r30, r31
	 moveli r32, _TIF_SYSCALL_ENTRY_WORK
	}
	and     r30, r30, r32
	{
	 addi   r30, r31, THREAD_INFO_STATUS_OFFSET - THREAD_INFO_FLAGS_OFFSET
	 beqzt	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:
	{
	 ld     r30, r30
	 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)
	{
	 ld     TREG_SYSCALL_NR_NAME, r11
	 moveli r21, __NR_syscalls
	}

	/* Ensure that the syscall number is within the legal range. */
	{
	 moveli r20, hw2(sys_call_table)
#ifdef CONFIG_COMPAT
	 blbs   r30, .Lcompat_syscall
#endif
	}
	{
	 cmpltu r21, TREG_SYSCALL_NR_NAME, r21
	 shl16insli r20, r20, hw1(sys_call_table)
	}
	{
	 blbc   r21, .Linvalid_syscall
	 shl16insli r20, r20, hw0(sys_call_table)
	}
.Lload_syscall_pointer:
	shl3add r20, TREG_SYSCALL_NR_NAME, r20
	ld      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.
	 * In compat mode, sign-extend r0 before storing it.
	 */
	{
	 PTREGS_PTR(r29, PTREGS_OFFSET_REG(0))
	 blbct  r30, 1f
	}
	addxi   r0, r0, 0
1:	st      r29, r0

.Lsyscall_sigreturn_skip:
	FEEDBACK_REENTER(handle_syscall)

	/* Do syscall trace again, if requested. */
	{
	 ld      r30, r31
	 moveli  r32, _TIF_SYSCALL_EXIT_WORK
	}
	and      r0, r30, r32
	{
	 andi    r0, r30, _TIF_SINGLESTEP
	 beqzt   r0, 1f
	}
	{
	 PTREGS_PTR(r0, PTREGS_OFFSET_BASE)
	 jal    do_syscall_trace_exit
	}
	FEEDBACK_REENTER(handle_syscall)
	andi    r0, r30, _TIF_SINGLESTEP

1:	beqzt	r0, 2f

	/* Single stepping -- notify ptrace. */
	{
	 movei   r0, SIGTRAP
	 jal     ptrace_notify
	}
	FEEDBACK_REENTER(handle_syscall)

2:	{
	 movei  r30, 0               /* not an NMI */
	 j      .Lresume_userspace   /* jump into middle of interrupt_return */
	}

#ifdef CONFIG_COMPAT
.Lcompat_syscall:
	/*
	 * Load the base of the compat syscall table in r20, and
	 * range-check the syscall number (duplicated from 64-bit path).
	 * Sign-extend all the user's passed arguments to make them consistent.
	 * Also save the original "r(n)" values away in "r(11+n)" in
	 * case the syscall table entry wants to validate them.
	 */
	moveli  r20, hw2(compat_sys_call_table)
	{
	 cmpltu r21, TREG_SYSCALL_NR_NAME, r21
	 shl16insli r20, r20, hw1(compat_sys_call_table)
	}
	{
	 blbc   r21, .Linvalid_syscall
	 shl16insli r20, r20, hw0(compat_sys_call_table)
	}
	{ move r11, r0; addxi r0, r0, 0 }
	{ move r12, r1; addxi r1, r1, 0 }
	{ move r13, r2; addxi r2, r2, 0 }
	{ move r14, r3; addxi r3, r3, 0 }
	{ move r15, r4; addxi r4, r4, 0 }
	{ move r16, r5; addxi r5, r5, 0 }
	j .Lload_syscall_pointer
#endif

.Linvalid_syscall:
	/* Report an invalid syscall back to the user program */
	{
	 PTREGS_PTR(r29, PTREGS_OFFSET_REG(0))
	 movei  r28, -ENOSYS
	}
	st      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)

/* 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)
#ifdef CONFIG_COMPAT
PTREGS_SYSCALL_SIGRETURN(compat_sys_rt_sigreturn, r0)
#endif

/* 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)

	/*
	 * Recover r3, r2, r1 and r0 here saved by unalign fast vector.
	 * The vector area limit is 32 bundles, so we handle the reload here.
	 * r0, r1, r2 are in thread_info from low to high memory in order.
	 * r3 points to location the original r3 was saved.
	 * We put this code in the __HEAD section so it can be reached
	 * via a conditional branch from the fast path.
	 */
	__HEAD
hand_unalign_slow:
	andi    sp, sp, ~1
hand_unalign_slow_badsp:
	addi    r3, r3, -(3 * 8)
	ld_add  r0, r3, 8
	ld_add  r1, r3, 8
	ld      r2, r3
hand_unalign_slow_nonuser:
	mfspr   r3, SPR_SYSTEM_SAVE_K_1
	__int_hand     INT_UNALIGN_DATA, UNALIGN_DATA_SLOW, int_unalign

/* The unaligned data support needs to read all the registers. */
int_unalign:
	push_extra_callee_saves r0
	j       do_unaligned
ENDPROC(hand_unalign_slow)

/* Fill the return address stack with nonzero entries. */
STD_ENTRY(fill_ra_stack)
	{
	 move	r0, lr
	 jal	1f
	}
1:	jal	2f
2:	jal	3f
3:	jal	4f
4:	jrp	r0
	STD_ENDPROC(fill_ra_stack)

	.macro int_hand  vecnum, vecname, c_routine, processing=handle_interrupt
	.org   (\vecnum << 8)
		__int_hand   \vecnum, \vecname, \c_routine, \processing
	.endm

/* Include .intrpt array of interrupt vectors */
	.section ".intrpt", "ax"
	.global intrpt_start
intrpt_start:

#ifndef CONFIG_USE_PMC
#define handle_perf_interrupt bad_intr
#endif

#ifndef CONFIG_HARDWALL
#define do_hardwall_trap bad_intr
#endif

	int_hand     INT_MEM_ERROR, MEM_ERROR, do_trap
	int_hand     INT_SINGLE_STEP_3, SINGLE_STEP_3, bad_intr
#if CONFIG_KERNEL_PL == 2
	int_hand     INT_SINGLE_STEP_2, SINGLE_STEP_2, gx_singlestep_handle
	int_hand     INT_SINGLE_STEP_1, SINGLE_STEP_1, bad_intr
#else
	int_hand     INT_SINGLE_STEP_2, SINGLE_STEP_2, bad_intr
	int_hand     INT_SINGLE_STEP_1, SINGLE_STEP_1, gx_singlestep_handle
#endif
	int_hand     INT_SINGLE_STEP_0, SINGLE_STEP_0, bad_intr
	int_hand     INT_IDN_COMPLETE, IDN_COMPLETE, bad_intr
	int_hand     INT_UDN_COMPLETE, UDN_COMPLETE, bad_intr
	int_hand     INT_ITLB_MISS, ITLB_MISS, do_page_fault
	int_hand     INT_ILL, ILL, do_trap
	int_hand     INT_GPV, GPV, do_trap
	int_hand     INT_IDN_ACCESS, IDN_ACCESS, do_trap
	int_hand     INT_UDN_ACCESS, UDN_ACCESS, do_trap
	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_ILL_TRANS, ILL_TRANS, do_trap
	int_hand_unalign_fast INT_UNALIGN_DATA, UNALIGN_DATA
	int_hand     INT_DTLB_MISS, DTLB_MISS, do_page_fault
	int_hand     INT_DTLB_ACCESS, DTLB_ACCESS, do_page_fault
	int_hand     INT_IDN_FIREWALL, IDN_FIREWALL, do_hardwall_trap
	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_IDN_AVAIL, IDN_AVAIL, bad_intr
	int_hand     INT_UDN_AVAIL, UDN_AVAIL, bad_intr
	int_hand     INT_IPI_3, IPI_3, bad_intr
#if CONFIG_KERNEL_PL == 2
	int_hand     INT_IPI_2, IPI_2, tile_dev_intr
	int_hand     INT_IPI_1, IPI_1, bad_intr
#else
	int_hand     INT_IPI_2, IPI_2, bad_intr
	int_hand     INT_IPI_1, IPI_1, tile_dev_intr
#endif
	int_hand     INT_IPI_0, IPI_0, bad_intr
	int_hand     INT_PERF_COUNT, PERF_COUNT, \
		     handle_perf_interrupt, handle_nmi
	int_hand     INT_AUX_PERF_COUNT, AUX_PERF_COUNT, \
		     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, bad_intr
	int_hand     INT_I_ASID, I_ASID, bad_intr
	int_hand     INT_D_ASID, D_ASID, bad_intr
	int_hand     INT_DOUBLE_FAULT, DOUBLE_FAULT, do_trap

	/* Synthetic interrupt delivered only by the simulator */
	int_hand     INT_BREAKPOINT, BREAKPOINT, do_breakpoint
	/* Synthetic interrupt delivered by hv */
	int_hand     INT_NMI_DWNCL, NMI_DWNCL, do_nmi, handle_nmi