aboutsummaryrefslogtreecommitdiff
path: root/boehm-gc/pthread_support.c
blob: 57c8a16a1fa574bfc5be282f7666d867fc1979c4 (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
/* 
 * Copyright (c) 1994 by Xerox Corporation.  All rights reserved.
 * Copyright (c) 1996 by Silicon Graphics.  All rights reserved.
 * Copyright (c) 1998 by Fergus Henderson.  All rights reserved.
 * Copyright (c) 2000-2004 by Hewlett-Packard Company.  All rights reserved.
 *
 * THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
 * OR IMPLIED.  ANY USE IS AT YOUR OWN RISK.
 *
 * Permission is hereby granted to use or copy this program
 * for any purpose,  provided the above notices are retained on all copies.
 * Permission to modify the code and to distribute modified code is granted,
 * provided the above notices are retained, and a notice that the code was
 * modified is included with the above copyright notice.
 */
/*
 * Support code for LinuxThreads, the clone()-based kernel
 * thread package for Linux which is included in libc6.
 *
 * This code relies on implementation details of LinuxThreads,
 * (i.e. properties not guaranteed by the Pthread standard),
 * though this version now does less of that than the other Pthreads
 * support code.
 *
 * Note that there is a lot of code duplication between linux_threads.c
 * and thread support for some of the other Posix platforms; any changes
 * made here may need to be reflected there too.
 */
 /* DG/UX ix86 support <takis@xfree86.org> */
/*
 * Linux_threads.c now also includes some code to support HPUX and
 * OSF1 (Compaq Tru64 Unix, really).  The OSF1 support is based on Eric Benson's
 * patch.
 *
 * Eric also suggested an alternate basis for a lock implementation in
 * his code:
 * + #elif defined(OSF1)
 * +    unsigned long GC_allocate_lock = 0;
 * +    msemaphore GC_allocate_semaphore;
 * + #  define GC_TRY_LOCK() \
 * +    ((msem_lock(&GC_allocate_semaphore, MSEM_IF_NOWAIT) == 0) \
 * +     ? (GC_allocate_lock = 1) \
 * +     : 0)
 * + #  define GC_LOCK_TAKEN GC_allocate_lock
 */

/*#define DEBUG_THREADS 1*/
/*#define GC_ASSERTIONS*/

#include "gc_config.h"

#ifdef GC_PTHREAD_SYM_VERSION
#define _GNU_SOURCE
#include <dlfcn.h>
#endif

# include "gc.h"
# include "private/pthread_support.h"

# if defined(GC_PTHREADS) && !defined(GC_SOLARIS_THREADS) \
     && !defined(GC_WIN32_THREADS)

# if defined(GC_HPUX_THREADS) && !defined(USE_PTHREAD_SPECIFIC) \
     && !defined(USE_COMPILER_TLS)
#   ifdef __GNUC__
#     define USE_PTHREAD_SPECIFIC
      /* Empirically, as of gcc 3.3, USE_COMPILER_TLS doesn't work.	*/
#   else
#     define USE_COMPILER_TLS
#   endif
# endif

# if defined USE_HPUX_TLS
    --> Macro replaced by USE_COMPILER_TLS
# endif

# if (defined(GC_DGUX386_THREADS) || defined(GC_OSF1_THREADS) || \
      defined(GC_DARWIN_THREADS) || defined(GC_AIX_THREADS)) \
      && !defined(USE_PTHREAD_SPECIFIC)
#   define USE_PTHREAD_SPECIFIC
# endif

# if defined(GC_DGUX386_THREADS) && !defined(_POSIX4A_DRAFT10_SOURCE)
#   define _POSIX4A_DRAFT10_SOURCE 1
# endif

# if defined(GC_DGUX386_THREADS) && !defined(_USING_POSIX4A_DRAFT10)
#   define _USING_POSIX4A_DRAFT10 1
# endif

# ifdef THREAD_LOCAL_ALLOC
#   if !defined(USE_PTHREAD_SPECIFIC) && !defined(USE_COMPILER_TLS)
#     include "private/specific.h"
#   endif
#   if defined(USE_PTHREAD_SPECIFIC)
#     define GC_getspecific pthread_getspecific
#     define GC_setspecific pthread_setspecific
#     define GC_key_create pthread_key_create
      typedef pthread_key_t GC_key_t;
#   endif
#   if defined(USE_COMPILER_TLS)
#     define GC_getspecific(x) (x)
#     define GC_setspecific(key, v) ((key) = (v), 0)
#     define GC_key_create(key, d) 0
      typedef void * GC_key_t;
#   endif
# endif
# include <stdlib.h>
# include <pthread.h>
# include <sched.h>
# include <time.h>
# include <errno.h>
# include <unistd.h>
# include <sys/mman.h>
# include <sys/time.h>
# include <sys/types.h>
# include <sys/stat.h>
# include <fcntl.h>
# include <signal.h>

#if defined(GC_DARWIN_THREADS)
# include "private/darwin_semaphore.h"
#else
# include <semaphore.h>
#endif /* !GC_DARWIN_THREADS */

#if defined(GC_DARWIN_THREADS) || defined(GC_FREEBSD_THREADS)
# include <sys/sysctl.h>
#endif /* GC_DARWIN_THREADS */



#if defined(GC_DGUX386_THREADS)
# include <sys/dg_sys_info.h>
# include <sys/_int_psem.h>
  /* sem_t is an uint in DG/UX */
  typedef unsigned int  sem_t;
#endif /* GC_DGUX386_THREADS */

#ifndef __GNUC__
#   define __inline__
#endif

#ifdef GC_USE_LD_WRAP
#   define WRAP_FUNC(f) __wrap_##f
#   define REAL_FUNC(f) __real_##f
#else
#   define WRAP_FUNC(f) GC_##f
#   if !defined(GC_DGUX386_THREADS)
#     define REAL_FUNC(f) f
#   else /* GC_DGUX386_THREADS */
#     define REAL_FUNC(f) __d10_##f
#   endif /* GC_DGUX386_THREADS */
#   undef pthread_create
#   if !defined(GC_DARWIN_THREADS)
#     undef pthread_sigmask
#   endif
#   undef pthread_join
#   undef pthread_detach
#   if defined(GC_OSF1_THREADS) && defined(_PTHREAD_USE_MANGLED_NAMES_) \
       && !defined(_PTHREAD_USE_PTDNAM_)
/* Restore the original mangled names on Tru64 UNIX.  */
#     define pthread_create __pthread_create
#     define pthread_join __pthread_join
#     define pthread_detach __pthread_detach
#   endif
#endif

void GC_thr_init();

static GC_bool parallel_initialized = FALSE;

void GC_init_parallel();

# if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)

/* We don't really support thread-local allocation with DBG_HDRS_ALL */

#ifdef USE_COMPILER_TLS
  __thread
#endif
GC_key_t GC_thread_key;

static GC_bool keys_initialized;

/* Recover the contents of the freelist array fl into the global one gfl.*/
/* Note that the indexing scheme differs, in that gfl has finer size	*/
/* resolution, even if not all entries are used.			*/
/* We hold the allocator lock.						*/
static void return_freelists(ptr_t *fl, ptr_t *gfl)
{
    int i;
    ptr_t q, *qptr;
    size_t nwords;

    for (i = 1; i < NFREELISTS; ++i) {
	nwords = i * (GRANULARITY/sizeof(word));
        qptr = fl + i;	
	q = *qptr;
	if ((word)q >= HBLKSIZE) {
	  if (gfl[nwords] == 0) {
	    gfl[nwords] = q;
	  } else {
	    /* Concatenate: */
	    for (; (word)q >= HBLKSIZE; qptr = &(obj_link(q)), q = *qptr);
	    GC_ASSERT(0 == q);
	    *qptr = gfl[nwords];
	    gfl[nwords] = fl[i];
	  }
	}
	/* Clear fl[i], since the thread structure may hang around.	*/
	/* Do it in a way that is likely to trap if we access it.	*/
	fl[i] = (ptr_t)HBLKSIZE;
    }
}

/* We statically allocate a single "size 0" object. It is linked to	*/
/* itself, and is thus repeatedly reused for all size 0 allocation	*/
/* requests.  (Size 0 gcj allocation requests are incorrect, and	*/
/* we arrange for those to fault asap.)					*/
static ptr_t size_zero_object = (ptr_t)(&size_zero_object);

/* Each thread structure must be initialized.	*/
/* This call must be made from the new thread.	*/
/* Caller holds allocation lock.		*/
void GC_init_thread_local(GC_thread p)
{
    int i;

    if (!keys_initialized) {
	if (0 != GC_key_create(&GC_thread_key, 0)) {
	    ABORT("Failed to create key for local allocator");
        }
	keys_initialized = TRUE;
    }
    if (0 != GC_setspecific(GC_thread_key, p)) {
	ABORT("Failed to set thread specific allocation pointers");
    }
    for (i = 1; i < NFREELISTS; ++i) {
	p -> ptrfree_freelists[i] = (ptr_t)1;
	p -> normal_freelists[i] = (ptr_t)1;
#	ifdef GC_GCJ_SUPPORT
	  p -> gcj_freelists[i] = (ptr_t)1;
#	endif
    }   
    /* Set up the size 0 free lists.	*/
    p -> ptrfree_freelists[0] = (ptr_t)(&size_zero_object);
    p -> normal_freelists[0] = (ptr_t)(&size_zero_object);
#   ifdef GC_GCJ_SUPPORT
        p -> gcj_freelists[0] = (ptr_t)(-1);
#   endif
}

#ifdef GC_GCJ_SUPPORT
  extern ptr_t * GC_gcjobjfreelist;
#endif

/* We hold the allocator lock.	*/
void GC_destroy_thread_local(GC_thread p)
{
    /* We currently only do this from the thread itself or from	*/
    /* the fork handler for a child process.			*/
#   ifndef HANDLE_FORK
      GC_ASSERT(GC_getspecific(GC_thread_key) == (void *)p);
#   endif
    return_freelists(p -> ptrfree_freelists, GC_aobjfreelist);
    return_freelists(p -> normal_freelists, GC_objfreelist);
#   ifdef GC_GCJ_SUPPORT
   	return_freelists(p -> gcj_freelists, GC_gcjobjfreelist);
#   endif
}

extern GC_PTR GC_generic_malloc_many();

GC_PTR GC_local_malloc(size_t bytes)
{
    if (EXPECT(!SMALL_ENOUGH(bytes),0)) {
        return(GC_malloc(bytes));
    } else {
	int index = INDEX_FROM_BYTES(bytes);
	ptr_t * my_fl;
	ptr_t my_entry;
#	if defined(REDIRECT_MALLOC) && !defined(USE_PTHREAD_SPECIFIC)
	GC_key_t k = GC_thread_key;
#	endif
	void * tsd;

#	if defined(REDIRECT_MALLOC) && !defined(USE_PTHREAD_SPECIFIC)
	    if (EXPECT(0 == k, 0)) {
		/* This can happen if we get called when the world is	*/
		/* being initialized.  Whether we can actually complete	*/
		/* the initialization then is unclear.			*/
		GC_init_parallel();
		k = GC_thread_key;
	    }
#	endif
	tsd = GC_getspecific(GC_thread_key);
#	ifdef GC_ASSERTIONS
	  LOCK();
	  GC_ASSERT(tsd == (void *)GC_lookup_thread(pthread_self()));
	  UNLOCK();
#	endif
	my_fl = ((GC_thread)tsd) -> normal_freelists + index;
	my_entry = *my_fl;
	if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
	    ptr_t next = obj_link(my_entry);
	    GC_PTR result = (GC_PTR)my_entry;
	    *my_fl = next;
	    obj_link(my_entry) = 0;
	    PREFETCH_FOR_WRITE(next);
	    return result;
	} else if ((word)my_entry - 1 < DIRECT_GRANULES) {
	    *my_fl = my_entry + index + 1;
            return GC_malloc(bytes);
	} else {
	    GC_generic_malloc_many(BYTES_FROM_INDEX(index), NORMAL, my_fl);
	    if (*my_fl == 0) return GC_oom_fn(bytes);
	    return GC_local_malloc(bytes);
	}
    }
}

GC_PTR GC_local_malloc_atomic(size_t bytes)
{
    if (EXPECT(!SMALL_ENOUGH(bytes), 0)) {
        return(GC_malloc_atomic(bytes));
    } else {
	int index = INDEX_FROM_BYTES(bytes);
	ptr_t * my_fl = ((GC_thread)GC_getspecific(GC_thread_key))
		        -> ptrfree_freelists + index;
	ptr_t my_entry = *my_fl;
    
	if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
	    GC_PTR result = (GC_PTR)my_entry;
	    *my_fl = obj_link(my_entry);
	    return result;
	} else if ((word)my_entry - 1 < DIRECT_GRANULES) {
	    *my_fl = my_entry + index + 1;
        return GC_malloc_atomic(bytes);
	} else {
	    GC_generic_malloc_many(BYTES_FROM_INDEX(index), PTRFREE, my_fl);
	    /* *my_fl is updated while the collector is excluded;	*/
	    /* the free list is always visible to the collector as 	*/
	    /* such.							*/
	    if (*my_fl == 0) return GC_oom_fn(bytes);
	    return GC_local_malloc_atomic(bytes);
	}
    }
}

#ifdef GC_GCJ_SUPPORT

#include "include/gc_gcj.h"

#ifdef GC_ASSERTIONS
  extern GC_bool GC_gcj_malloc_initialized;
#endif

extern int GC_gcj_kind;

GC_PTR GC_local_gcj_malloc(size_t bytes,
			   void * ptr_to_struct_containing_descr)
{
    GC_ASSERT(GC_gcj_malloc_initialized);
    if (EXPECT(!SMALL_ENOUGH(bytes), 0)) {
        return GC_gcj_malloc(bytes, ptr_to_struct_containing_descr);
    } else {
	int index = INDEX_FROM_BYTES(bytes);
	ptr_t * my_fl = ((GC_thread)GC_getspecific(GC_thread_key))
	                -> gcj_freelists + index;
	ptr_t my_entry = *my_fl;
	if (EXPECT((word)my_entry >= HBLKSIZE, 1)) {
	    GC_PTR result = (GC_PTR)my_entry;
	    GC_ASSERT(!GC_incremental);
	    /* We assert that any concurrent marker will stop us.	*/
	    /* Thus it is impossible for a mark procedure to see the 	*/
	    /* allocation of the next object, but to see this object 	*/
	    /* still containing a free list pointer.  Otherwise the 	*/
	    /* marker might find a random "mark descriptor".		*/
	    *(volatile ptr_t *)my_fl = obj_link(my_entry);
	    /* We must update the freelist before we store the pointer.	*/
	    /* Otherwise a GC at this point would see a corrupted	*/
	    /* free list.						*/
	    /* A memory barrier is probably never needed, since the 	*/
	    /* action of stopping this thread will cause prior writes	*/
	    /* to complete.						*/
	    GC_ASSERT(((void * volatile *)result)[1] == 0); 
	    *(void * volatile *)result = ptr_to_struct_containing_descr; 
	    return result;
	} else if ((word)my_entry - 1 < DIRECT_GRANULES) {
	    if (!GC_incremental) *my_fl = my_entry + index + 1;
	    	/* In the incremental case, we always have to take this */
	    	/* path.  Thus we leave the counter alone.		*/
            return GC_gcj_malloc(bytes, ptr_to_struct_containing_descr);
	} else {
	    GC_generic_malloc_many(BYTES_FROM_INDEX(index), GC_gcj_kind, my_fl);
	    if (*my_fl == 0) return GC_oom_fn(bytes);
	    return GC_local_gcj_malloc(bytes, ptr_to_struct_containing_descr);
	}
    }
}

#endif /* GC_GCJ_SUPPORT */

# else  /* !THREAD_LOCAL_ALLOC  && !DBG_HDRS_ALL */

#   define GC_destroy_thread_local(t)

# endif /* !THREAD_LOCAL_ALLOC */

#if 0
/*
To make sure that we're using LinuxThreads and not some other thread
package, we generate a dummy reference to `pthread_kill_other_threads_np'
(was `__pthread_initial_thread_bos' but that disappeared),
which is a symbol defined in LinuxThreads, but (hopefully) not in other
thread packages.

We no longer do this, since this code is now portable enough that it might
actually work for something else.
*/
void (*dummy_var_to_force_linux_threads)() = pthread_kill_other_threads_np;
#endif /* 0 */

long GC_nprocs = 1;	/* Number of processors.  We may not have	*/
			/* access to all of them, but this is as good	*/
			/* a guess as any ...				*/

#ifdef PARALLEL_MARK

# ifndef MAX_MARKERS
#   define MAX_MARKERS 16
# endif

static ptr_t marker_sp[MAX_MARKERS] = {0};

void * GC_mark_thread(void * id)
{
  word my_mark_no = 0;

  marker_sp[(word)id] = GC_approx_sp();
  for (;; ++my_mark_no) {
    /* GC_mark_no is passed only to allow GC_help_marker to terminate	*/
    /* promptly.  This is important if it were called from the signal	*/
    /* handler or from the GC lock acquisition code.  Under Linux, it's	*/
    /* not safe to call it from a signal handler, since it uses mutexes	*/
    /* and condition variables.  Since it is called only here, the 	*/
    /* argument is unnecessary.						*/
    if (my_mark_no < GC_mark_no || my_mark_no > GC_mark_no + 2) {
	/* resynchronize if we get far off, e.g. because GC_mark_no	*/
	/* wrapped.							*/
	my_mark_no = GC_mark_no;
    }
#   ifdef DEBUG_THREADS
	GC_printf1("Starting mark helper for mark number %ld\n", my_mark_no);
#   endif
    GC_help_marker(my_mark_no);
  }
}

extern long GC_markers;		/* Number of mark threads we would	*/
				/* like to have.  Includes the 		*/
				/* initiating thread.			*/

pthread_t GC_mark_threads[MAX_MARKERS];

#define PTHREAD_CREATE REAL_FUNC(pthread_create)

static void start_mark_threads()
{
    unsigned i;
    pthread_attr_t attr;

    if (GC_markers > MAX_MARKERS) {
	WARN("Limiting number of mark threads\n", 0);
	GC_markers = MAX_MARKERS;
    }
    if (0 != pthread_attr_init(&attr)) ABORT("pthread_attr_init failed");
	
    if (0 != pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED))
	ABORT("pthread_attr_setdetachstate failed");

#   if defined(HPUX) || defined(GC_DGUX386_THREADS)
      /* Default stack size is usually too small: fix it. */
      /* Otherwise marker threads or GC may run out of	  */
      /* space.						  */
#     define MIN_STACK_SIZE (8*HBLKSIZE*sizeof(word))
      {
	size_t old_size;
	int code;

        if (pthread_attr_getstacksize(&attr, &old_size) != 0)
	  ABORT("pthread_attr_getstacksize failed\n");
	if (old_size < MIN_STACK_SIZE) {
	  if (pthread_attr_setstacksize(&attr, MIN_STACK_SIZE) != 0)
		  ABORT("pthread_attr_setstacksize failed\n");
	}
      }
#   endif /* HPUX || GC_DGUX386_THREADS */
#   ifdef CONDPRINT
      if (GC_print_stats) {
	GC_printf1("Starting %ld marker threads\n", GC_markers - 1);
      }
#   endif
    for (i = 0; i < GC_markers - 1; ++i) {
      if (0 != PTHREAD_CREATE(GC_mark_threads + i, &attr,
			      GC_mark_thread, (void *)(word)i)) {
	WARN("Marker thread creation failed, errno = %ld.\n", errno);
      }
    }
}

#else  /* !PARALLEL_MARK */

static __inline__ void start_mark_threads()
{
}

#endif /* !PARALLEL_MARK */

GC_bool GC_thr_initialized = FALSE;

volatile GC_thread GC_threads[THREAD_TABLE_SZ];

void GC_push_thread_structures GC_PROTO((void))
{
    GC_push_all((ptr_t)(GC_threads), (ptr_t)(GC_threads)+sizeof(GC_threads));
#   if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
      GC_push_all((ptr_t)(&GC_thread_key),
	  (ptr_t)(&GC_thread_key)+sizeof(&GC_thread_key));
#   endif
}

#ifdef THREAD_LOCAL_ALLOC
/* We must explicitly mark ptrfree and gcj free lists, since the free 	*/
/* list links wouldn't otherwise be found.  We also set them in the 	*/
/* normal free lists, since that involves touching less memory than if	*/
/* we scanned them normally.						*/
void GC_mark_thread_local_free_lists(void)
{
    int i, j;
    GC_thread p;
    ptr_t q;
    
    for (i = 0; i < THREAD_TABLE_SZ; ++i) {
      for (p = GC_threads[i]; 0 != p; p = p -> next) {
	for (j = 1; j < NFREELISTS; ++j) {
	  q = p -> ptrfree_freelists[j];
	  if ((word)q > HBLKSIZE) GC_set_fl_marks(q);
	  q = p -> normal_freelists[j];
	  if ((word)q > HBLKSIZE) GC_set_fl_marks(q);
#	  ifdef GC_GCJ_SUPPORT
	    q = p -> gcj_freelists[j];
	    if ((word)q > HBLKSIZE) GC_set_fl_marks(q);
#	  endif /* GC_GCJ_SUPPORT */
	}
      }
    }
}
#endif /* THREAD_LOCAL_ALLOC */

static struct GC_Thread_Rep first_thread;

/* Add a thread to GC_threads.  We assume it wasn't already there.	*/
/* Caller holds allocation lock.					*/
GC_thread GC_new_thread(pthread_t id)
{
    int hv = ((word)id) % THREAD_TABLE_SZ;
    GC_thread result;
    static GC_bool first_thread_used = FALSE;
    
    if (!first_thread_used) {
    	result = &first_thread;
    	first_thread_used = TRUE;
    } else {
        result = (struct GC_Thread_Rep *)
        	 GC_INTERNAL_MALLOC(sizeof(struct GC_Thread_Rep), NORMAL);
    }
    if (result == 0) return(0);
    result -> id = id;
    result -> next = GC_threads[hv];
    GC_threads[hv] = result;
    GC_ASSERT(result -> flags == 0 && result -> thread_blocked == 0);
    return(result);
}

/* Delete a thread from GC_threads.  We assume it is there.	*/
/* (The code intentionally traps if it wasn't.)			*/
/* Caller holds allocation lock.				*/
void GC_delete_thread(pthread_t id)
{
    int hv = ((word)id) % THREAD_TABLE_SZ;
    register GC_thread p = GC_threads[hv];
    register GC_thread prev = 0;
    
    while (!pthread_equal(p -> id, id)) {
        prev = p;
        p = p -> next;
    }
    if (prev == 0) {
        GC_threads[hv] = p -> next;
    } else {
        prev -> next = p -> next;
    }

    if (p != &first_thread)
      GC_INTERNAL_FREE(p);
}

/* If a thread has been joined, but we have not yet		*/
/* been notified, then there may be more than one thread 	*/
/* in the table with the same pthread id.			*/
/* This is OK, but we need a way to delete a specific one.	*/
void GC_delete_gc_thread(pthread_t id, GC_thread gc_id)
{
    int hv = ((word)id) % THREAD_TABLE_SZ;
    register GC_thread p = GC_threads[hv];
    register GC_thread prev = 0;

    while (p != gc_id) {
        prev = p;
        p = p -> next;
    }
    if (prev == 0) {
        GC_threads[hv] = p -> next;
    } else {
        prev -> next = p -> next;
    }
    GC_INTERNAL_FREE(p);
}

/* Return a GC_thread corresponding to a given pthread_t.	*/
/* Returns 0 if it's not there.					*/
/* Caller holds  allocation lock or otherwise inhibits 		*/
/* updates.							*/
/* If there is more than one thread with the given id we 	*/
/* return the most recent one.					*/
GC_thread GC_lookup_thread(pthread_t id)
{
    int hv = ((word)id) % THREAD_TABLE_SZ;
    register GC_thread p = GC_threads[hv];
    
    while (p != 0 && !pthread_equal(p -> id, id)) p = p -> next;
    return(p);
}

#ifdef HANDLE_FORK
/* Remove all entries from the GC_threads table, except the	*/
/* one for the current thread.  We need to do this in the child	*/
/* process after a fork(), since only the current thread 	*/
/* survives in the child.					*/
void GC_remove_all_threads_but_me(void)
{
    pthread_t self = pthread_self();
    int hv;
    GC_thread p, next, me;

    for (hv = 0; hv < THREAD_TABLE_SZ; ++hv) {
      me = 0;
      for (p = GC_threads[hv]; 0 != p; p = next) {
	next = p -> next;
	if (p -> id == self) {
	  me = p;
	  p -> next = 0;
	} else {
#	  ifdef THREAD_LOCAL_ALLOC
	    if (!(p -> flags & FINISHED)) {
	      GC_destroy_thread_local(p);
	    }
#	  endif /* THREAD_LOCAL_ALLOC */
	  if (p != &first_thread) GC_INTERNAL_FREE(p);
	}
      }
      GC_threads[hv] = me;
    }
}
#endif /* HANDLE_FORK */

#ifdef USE_PROC_FOR_LIBRARIES
int GC_segment_is_thread_stack(ptr_t lo, ptr_t hi)
{
    int i;
    GC_thread p;
    
#   ifdef PARALLEL_MARK
      for (i = 0; i < GC_markers; ++i) {
	if (marker_sp[i] > lo & marker_sp[i] < hi) return 1;
      }
#   endif
    for (i = 0; i < THREAD_TABLE_SZ; i++) {
      for (p = GC_threads[i]; p != 0; p = p -> next) {
	if (0 != p -> stack_end) {
#	  ifdef STACK_GROWS_UP
            if (p -> stack_end >= lo && p -> stack_end < hi) return 1;
#	  else /* STACK_GROWS_DOWN */
            if (p -> stack_end > lo && p -> stack_end <= hi) return 1;
#	  endif
	}
      }
    }
    return 0;
}
#endif /* USE_PROC_FOR_LIBRARIES */

#ifdef GC_LINUX_THREADS
/* Return the number of processors, or i<= 0 if it can't be determined.	*/
int GC_get_nprocs()
{
    /* Should be "return sysconf(_SC_NPROCESSORS_ONLN);" but that	*/
    /* appears to be buggy in many cases.				*/
    /* We look for lines "cpu<n>" in /proc/stat.			*/
#   define STAT_BUF_SIZE 4096
#   define STAT_READ read
	/* If read is wrapped, this may need to be redefined to call 	*/
	/* the real one.						*/
    char stat_buf[STAT_BUF_SIZE];
    int f;
    word result = 1;
	/* Some old kernels only have a single "cpu nnnn ..."	*/
	/* entry in /proc/stat.  We identify those as 		*/
	/* uniprocessors.					*/
    size_t i, len = 0;

    f = open("/proc/stat", O_RDONLY);
    if (f < 0 || (len = STAT_READ(f, stat_buf, STAT_BUF_SIZE)) < 100) {
	WARN("Couldn't read /proc/stat\n", 0);
	return -1;
    }
    for (i = 0; i < len - 100; ++i) {
        if (stat_buf[i] == '\n' && stat_buf[i+1] == 'c'
	    && stat_buf[i+2] == 'p' && stat_buf[i+3] == 'u') {
	    int cpu_no = atoi(stat_buf + i + 4);
	    if (cpu_no >= result) result = cpu_no + 1;
	}
    }
    close(f);
    return result;
}
#endif /* GC_LINUX_THREADS */

/* We hold the GC lock.  Wait until an in-progress GC has finished.	*/
/* Repeatedly RELEASES GC LOCK in order to wait.			*/
/* If wait_for_all is true, then we exit with the GC lock held and no	*/
/* collection in progress; otherwise we just wait for the current GC	*/
/* to finish.								*/
extern GC_bool GC_collection_in_progress();
void GC_wait_for_gc_completion(GC_bool wait_for_all)
{
    if (GC_incremental && GC_collection_in_progress()) {
	int old_gc_no = GC_gc_no;

	/* Make sure that no part of our stack is still on the mark stack, */
	/* since it's about to be unmapped.				   */
	while (GC_incremental && GC_collection_in_progress()
	       && (wait_for_all || old_gc_no == GC_gc_no)) {
	    ENTER_GC();
	    GC_in_thread_creation = TRUE;
            GC_collect_a_little_inner(1);
	    GC_in_thread_creation = FALSE;
	    EXIT_GC();
	    UNLOCK();
	    sched_yield();
	    LOCK();
	}
    }
}

#ifdef HANDLE_FORK
/* Procedures called before and after a fork.  The goal here is to make */
/* it safe to call GC_malloc() in a forked child.  It's unclear that is	*/
/* attainable, since the single UNIX spec seems to imply that one 	*/
/* should only call async-signal-safe functions, and we probably can't	*/
/* quite guarantee that.  But we give it our best shot.  (That same	*/
/* spec also implies that it's not safe to call the system malloc	*/
/* between fork() and exec().  Thus we're doing no worse than it.	*/

/* Called before a fork()		*/
void GC_fork_prepare_proc(void)
{
    /* Acquire all relevant locks, so that after releasing the locks	*/
    /* the child will see a consistent state in which monitor 		*/
    /* invariants hold.	 Unfortunately, we can't acquire libc locks	*/
    /* we might need, and there seems to be no guarantee that libc	*/
    /* must install a suitable fork handler.				*/
    /* Wait for an ongoing GC to finish, since we can't finish it in	*/
    /* the (one remaining thread in) the child.				*/
      LOCK();
#     if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
        GC_wait_for_reclaim();
#     endif
      GC_wait_for_gc_completion(TRUE);
#     if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
        GC_acquire_mark_lock();
#     endif
}

/* Called in parent after a fork()	*/
void GC_fork_parent_proc(void)
{
#   if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
      GC_release_mark_lock();
#   endif
    UNLOCK();
}

/* Called in child after a fork()	*/
void GC_fork_child_proc(void)
{
    /* Clean up the thread table, so that just our thread is left. */
#   if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)
      GC_release_mark_lock();
#   endif
    GC_remove_all_threads_but_me();
#   ifdef PARALLEL_MARK
      /* Turn off parallel marking in the child, since we are probably 	*/
      /* just going to exec, and we would have to restart mark threads.	*/
        GC_markers = 1;
        GC_parallel = FALSE;
#   endif /* PARALLEL_MARK */
    UNLOCK();
}
#endif /* HANDLE_FORK */

#if defined(GC_DGUX386_THREADS)
/* Return the number of processors, or i<= 0 if it can't be determined. */
int GC_get_nprocs()
{
    /* <takis@XFree86.Org> */
    int numCpus;
    struct dg_sys_info_pm_info pm_sysinfo;
    int status =0;

    status = dg_sys_info((long int *) &pm_sysinfo,
	DG_SYS_INFO_PM_INFO_TYPE, DG_SYS_INFO_PM_CURRENT_VERSION);
    if (status < 0)
       /* set -1 for error */
       numCpus = -1;
    else
      /* Active CPUs */
      numCpus = pm_sysinfo.idle_vp_count;

#  ifdef DEBUG_THREADS
    GC_printf1("Number of active CPUs in this system: %d\n", numCpus);
#  endif
    return(numCpus);
}
#endif /* GC_DGUX386_THREADS */

/* We hold the allocation lock.	*/
void GC_thr_init()
{
#   ifndef GC_DARWIN_THREADS
      int dummy;
#   endif
    GC_thread t;

    if (GC_thr_initialized) return;
    GC_thr_initialized = TRUE;
    
#   ifdef HANDLE_FORK
      /* Prepare for a possible fork.	*/
        pthread_atfork(GC_fork_prepare_proc, GC_fork_parent_proc,
	  	       GC_fork_child_proc);
#   endif /* HANDLE_FORK */
    /* Add the initial thread, so we can stop it.	*/
      t = GC_new_thread(pthread_self());
#     ifdef GC_DARWIN_THREADS
         t -> stop_info.mach_thread = mach_thread_self();
#     else
         t -> stop_info.stack_ptr = (ptr_t)(&dummy);
#     endif
      t -> flags = DETACHED | MAIN_THREAD;

    GC_stop_init();

    /* Set GC_nprocs.  */
      {
	char * nprocs_string = GETENV("GC_NPROCS");
	GC_nprocs = -1;
	if (nprocs_string != NULL) GC_nprocs = atoi(nprocs_string);
      }
      if (GC_nprocs <= 0) {
#       if defined(GC_HPUX_THREADS)
	  GC_nprocs = pthread_num_processors_np();
#       endif
#	if defined(GC_OSF1_THREADS) || defined(GC_AIX_THREADS) \
	   || defined(GC_SOLARIS_PTHREADS)
	  GC_nprocs = sysconf(_SC_NPROCESSORS_ONLN);
	  if (GC_nprocs <= 0) GC_nprocs = 1;
#	endif
#       if defined(GC_IRIX_THREADS)
	  GC_nprocs = sysconf(_SC_NPROC_ONLN);
	  if (GC_nprocs <= 0) GC_nprocs = 1;
#       endif
#       if defined(GC_DARWIN_THREADS) || defined(GC_FREEBSD_THREADS)
	  int ncpus = 1;
	  size_t len = sizeof(ncpus);
	  sysctl((int[2]) {CTL_HW, HW_NCPU}, 2, &ncpus, &len, NULL, 0);
	  GC_nprocs = ncpus;
#       endif
#	if defined(GC_LINUX_THREADS) || defined(GC_DGUX386_THREADS)
          GC_nprocs = GC_get_nprocs();
#	endif
      }
      if (GC_nprocs <= 0) {
	WARN("GC_get_nprocs() returned %ld\n", GC_nprocs);
	GC_nprocs = 2;
#	ifdef PARALLEL_MARK
	  GC_markers = 1;
#	endif
      } else {
#	ifdef PARALLEL_MARK
          {
	    char * markers_string = GETENV("GC_MARKERS");
	    if (markers_string != NULL) {
	      GC_markers = atoi(markers_string);
	    } else {
	      GC_markers = GC_nprocs;
	    }
          }
#	endif
      }
#   ifdef PARALLEL_MARK
#     ifdef CONDPRINT
        if (GC_print_stats) {
          GC_printf2("Number of processors = %ld, "
		 "number of marker threads = %ld\n", GC_nprocs, GC_markers);
	}
#     endif
      if (GC_markers == 1) {
	GC_parallel = FALSE;
#	ifdef CONDPRINT
	  if (GC_print_stats) {
	    GC_printf0("Single marker thread, turning off parallel marking\n");
	  }
#	endif
      } else {
	GC_parallel = TRUE;
	/* Disable true incremental collection, but generational is OK.	*/
	GC_time_limit = GC_TIME_UNLIMITED;
      }
      /* If we are using a parallel marker, actually start helper threads.  */
        if (GC_parallel) start_mark_threads();
#   endif
}


/* Perform all initializations, including those that	*/
/* may require allocation.				*/
/* Called without allocation lock.			*/
/* Must be called before a second thread is created.	*/
/* Called without allocation lock.			*/
void GC_init_parallel()
{
    if (parallel_initialized) return;
    parallel_initialized = TRUE;

    /* GC_init() calls us back, so set flag first.	*/
    if (!GC_is_initialized) GC_init();
    /* Initialize thread local free lists if used.	*/
#   if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
      LOCK();
      GC_init_thread_local(GC_lookup_thread(pthread_self()));
      UNLOCK();
#   endif
}


#if !defined(GC_DARWIN_THREADS)
int WRAP_FUNC(pthread_sigmask)(int how, const sigset_t *set, sigset_t *oset)
{
    sigset_t fudged_set;
    
    if (set != NULL && (how == SIG_BLOCK || how == SIG_SETMASK)) {
        fudged_set = *set;
        sigdelset(&fudged_set, SIG_SUSPEND);
        set = &fudged_set;
    }
    return(REAL_FUNC(pthread_sigmask)(how, set, oset));
}
#endif /* !GC_DARWIN_THREADS */

/* Wrappers for functions that are likely to block for an appreciable	*/
/* length of time.  Must be called in pairs, if at all.			*/
/* Nothing much beyond the system call itself should be executed	*/
/* between these.							*/

void GC_start_blocking(void) {
#   define SP_SLOP 128
    GC_thread me;
    LOCK();
    me = GC_lookup_thread(pthread_self());
    GC_ASSERT(!(me -> thread_blocked));
#   ifdef SPARC
	me -> stop_info.stack_ptr = (ptr_t)GC_save_regs_in_stack();
#   else
#   ifndef GC_DARWIN_THREADS
	me -> stop_info.stack_ptr = (ptr_t)GC_approx_sp();
#   endif
#   endif
#   ifdef IA64
	me -> backing_store_ptr = (ptr_t)GC_save_regs_in_stack() + SP_SLOP;
#   endif
    /* Add some slop to the stack pointer, since the wrapped call may 	*/
    /* end up pushing more callee-save registers.			*/
#   ifndef GC_DARWIN_THREADS
#   ifdef STACK_GROWS_UP
	me -> stop_info.stack_ptr += SP_SLOP;
#   else
	me -> stop_info.stack_ptr -= SP_SLOP;
#   endif
#   endif
    me -> thread_blocked = TRUE;
    UNLOCK();
}

void GC_end_blocking(void) {
    GC_thread me;
    LOCK();   /* This will block if the world is stopped.	*/
    me = GC_lookup_thread(pthread_self());
    GC_ASSERT(me -> thread_blocked);
    me -> thread_blocked = FALSE;
    UNLOCK();
}
    
#if defined(GC_DGUX386_THREADS)
#define __d10_sleep sleep
#endif /* GC_DGUX386_THREADS */

/* A wrapper for the standard C sleep function	*/
int WRAP_FUNC(sleep) (unsigned int seconds)
{
    int result;

    GC_start_blocking();
    result = REAL_FUNC(sleep)(seconds);
    GC_end_blocking();
    return result;
}

struct start_info {
    void *(*start_routine)(void *);
    void *arg;
    word flags;
    sem_t registered;   	/* 1 ==> in our thread table, but 	*/
				/* parent hasn't yet noticed.		*/
};

/* Called at thread exit.				*/
/* Never called for main thread.  That's OK, since it	*/
/* results in at most a tiny one-time leak.  And 	*/
/* linuxthreads doesn't reclaim the main threads 	*/
/* resources or id anyway.				*/
void GC_thread_exit_proc(void *arg)
{
    GC_thread me;

    LOCK();
    me = GC_lookup_thread(pthread_self());
    GC_destroy_thread_local(me);
    if (me -> flags & DETACHED) {
    	GC_delete_thread(pthread_self());
    } else {
	me -> flags |= FINISHED;
    }
#   if defined(THREAD_LOCAL_ALLOC) && !defined(USE_PTHREAD_SPECIFIC) \
       && !defined(USE_COMPILER_TLS) && !defined(DBG_HDRS_ALL)
      GC_remove_specific(GC_thread_key);
#   endif
    /* The following may run the GC from "nonexistent" thread.	*/
    GC_wait_for_gc_completion(FALSE);
    UNLOCK();
}

int WRAP_FUNC(pthread_join)(pthread_t thread, void **retval)
{
    int result;
    GC_thread thread_gc_id;
    
    LOCK();
    thread_gc_id = GC_lookup_thread(thread);
    /* This is guaranteed to be the intended one, since the thread id	*/
    /* cant have been recycled by pthreads.				*/
    UNLOCK();
    result = REAL_FUNC(pthread_join)(thread, retval);
# if defined (GC_FREEBSD_THREADS)
    /* On FreeBSD, the wrapped pthread_join() sometimes returns (what
       appears to be) a spurious EINTR which caused the test and real code
       to gratuitously fail.  Having looked at system pthread library source
       code, I see how this return code may be generated.  In one path of
       code, pthread_join() just returns the errno setting of the thread
       being joined.  This does not match the POSIX specification or the
       local man pages thus I have taken the liberty to catch this one
       spurious return value properly conditionalized on GC_FREEBSD_THREADS. */
    if (result == EINTR) result = 0;
# endif
    if (result == 0) {
        LOCK();
        /* Here the pthread thread id may have been recycled. */
        GC_delete_gc_thread(thread, thread_gc_id);
        UNLOCK();
    }
    return result;
}

int
WRAP_FUNC(pthread_detach)(pthread_t thread)
{
    int result;
    GC_thread thread_gc_id;
    
    LOCK();
    thread_gc_id = GC_lookup_thread(thread);
    UNLOCK();
    result = REAL_FUNC(pthread_detach)(thread);
    if (result == 0) {
      LOCK();
      thread_gc_id -> flags |= DETACHED;
      /* Here the pthread thread id may have been recycled. */
      if (thread_gc_id -> flags & FINISHED) {
        GC_delete_gc_thread(thread, thread_gc_id);
      }
      UNLOCK();
    }
    return result;
}

GC_bool GC_in_thread_creation = FALSE;

GC_PTR GC_get_thread_stack_base()
{  
# ifdef HAVE_PTHREAD_GETATTR_NP
  pthread_t my_pthread;
  pthread_attr_t attr;
  ptr_t stack_addr;
  size_t stack_size;
  
  my_pthread = pthread_self();  
  if (pthread_getattr_np (my_pthread, &attr) != 0)
    {
#   ifdef DEBUG_THREADS
      GC_printf1("Can not determine stack base for attached thread");
#   endif
      return 0;
    }
  pthread_attr_getstack (&attr, (void **) &stack_addr, &stack_size);
  pthread_attr_destroy (&attr);
  
#   ifdef DEBUG_THREADS
	GC_printf1("attached thread stack address: 0x%x\n", stack_addr);
#   endif

#   ifdef STACK_GROWS_DOWN
      return stack_addr + stack_size;
#   else
      return stack_addr;
#   endif

# else
#   ifdef DEBUG_THREADS
	GC_printf1("Can not determine stack base for attached thread");
#   endif
  return 0;
# endif
}

void GC_register_my_thread()
{
  GC_thread me;
  pthread_t my_pthread;

  my_pthread = pthread_self();
#   ifdef DEBUG_THREADS
      GC_printf1("Attaching thread 0x%lx\n", my_pthread);
      GC_printf1("pid = %ld\n", (long) getpid());
#   endif
  
  /* Check to ensure this thread isn't attached already. */
  LOCK();
  me = GC_lookup_thread (my_pthread);
  UNLOCK();
  if (me != 0)
    {
#   ifdef DEBUG_THREADS
      GC_printf1("Attempt to re-attach known thread 0x%lx\n", my_pthread);
#   endif
      return;
    }

  LOCK();
  GC_in_thread_creation = TRUE;
  me = GC_new_thread(my_pthread);
  GC_in_thread_creation = FALSE;

  me -> flags |= DETACHED;  

#ifdef GC_DARWIN_THREADS
    me -> stop_info.mach_thread = mach_thread_self();
#else
    me -> stack_end = GC_get_thread_stack_base();    
    if (me -> stack_end == 0)
      GC_abort("Can not determine stack base for attached thread");
    
#   ifdef STACK_GROWS_DOWN
      me -> stop_info.stack_ptr = me -> stack_end - 0x10;
#   else
      me -> stop_info.stack_ptr = me -> stack_end + 0x10;
#   endif
#endif

#   ifdef IA64
      me -> backing_store_end = (ptr_t)
			(GC_save_regs_in_stack() & ~(GC_page_size - 1));
      /* This is also < 100% convincing.  We should also read this 	*/
      /* from /proc, but the hook to do so isn't there yet.		*/
#   endif /* IA64 */

#   if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
        GC_init_thread_local(me);
#   endif
  UNLOCK();
}

void GC_unregister_my_thread()
{
  pthread_t my_pthread;

  my_pthread = pthread_self();

#   ifdef DEBUG_THREADS
      GC_printf1("Detaching thread 0x%lx\n", my_pthread);
#   endif

  GC_thread_exit_proc (0);
}

void * GC_start_routine(void * arg)
{
    int dummy;
    struct start_info * si = arg;
    void * result;
    GC_thread me;
    pthread_t my_pthread;
    void *(*start)(void *);
    void *start_arg;

    my_pthread = pthread_self();
#   ifdef DEBUG_THREADS
        GC_printf1("Starting thread 0x%lx\n", my_pthread);
        GC_printf1("pid = %ld\n", (long) getpid());
        GC_printf1("sp = 0x%lx\n", (long) &arg);
#   endif
    LOCK();
    GC_in_thread_creation = TRUE;
    me = GC_new_thread(my_pthread);
    GC_in_thread_creation = FALSE;
#ifdef GC_DARWIN_THREADS
    me -> stop_info.mach_thread = mach_thread_self();
#else
    me -> stop_info.stack_ptr = 0;
#endif
    me -> flags = si -> flags;
    /* me -> stack_end = GC_linux_stack_base(); -- currently (11/99)	*/
    /* doesn't work because the stack base in /proc/self/stat is the 	*/
    /* one for the main thread.  There is a strong argument that that's	*/
    /* a kernel bug, but a pervasive one.				*/
#   ifdef STACK_GROWS_DOWN
      me -> stack_end = (ptr_t)(((word)(&dummy) + (GC_page_size - 1))
		                & ~(GC_page_size - 1));
#	  ifndef GC_DARWIN_THREADS
        me -> stop_info.stack_ptr = me -> stack_end - 0x10;
#	  endif
	/* Needs to be plausible, since an asynchronous stack mark	*/
	/* should not crash.						*/
#   else
      me -> stack_end = (ptr_t)((word)(&dummy) & ~(GC_page_size - 1));
      me -> stop_info.stack_ptr = me -> stack_end + 0x10;
#   endif
    /* This is dubious, since we may be more than a page into the stack, */
    /* and hence skip some of it, though it's not clear that matters.	 */
#   ifdef IA64
      me -> backing_store_end = (ptr_t)
			(GC_save_regs_in_stack() & ~(GC_page_size - 1));
      /* This is also < 100% convincing.  We should also read this 	*/
      /* from /proc, but the hook to do so isn't there yet.		*/
#   endif /* IA64 */
    UNLOCK();
    start = si -> start_routine;
#   ifdef DEBUG_THREADS
	GC_printf1("start_routine = 0x%lx\n", start);
#   endif
    start_arg = si -> arg;
    sem_post(&(si -> registered));	/* Last action on si.	*/
    					/* OK to deallocate.	*/
    pthread_cleanup_push(GC_thread_exit_proc, 0);
#   if defined(THREAD_LOCAL_ALLOC) && !defined(DBG_HDRS_ALL)
 	LOCK();
        GC_init_thread_local(me);
	UNLOCK();
#   endif
    result = (*start)(start_arg);
#if DEBUG_THREADS
        GC_printf1("Finishing thread 0x%x\n", pthread_self());
#endif
    me -> status = result;
    pthread_cleanup_pop(1);
    /* Cleanup acquires lock, ensuring that we can't exit		*/
    /* while a collection that thinks we're alive is trying to stop     */
    /* us.								*/
    return(result);
}

int
WRAP_FUNC(pthread_create)(pthread_t *new_thread,
		  const pthread_attr_t *attr,
                  void *(*start_routine)(void *), void *arg)
{
    int result;
    int detachstate;
    word my_flags = 0;
    struct start_info * si; 
	/* This is otherwise saved only in an area mmapped by the thread */
	/* library, which isn't visible to the collector.		 */
 
    /* We resist the temptation to muck with the stack size here,	*/
    /* even if the default is unreasonably small.  That's the client's	*/
    /* responsibility.							*/

    LOCK();
    si = (struct start_info *)GC_INTERNAL_MALLOC(sizeof(struct start_info),
						 NORMAL);
    UNLOCK();
    if (!parallel_initialized) GC_init_parallel();
    if (0 == si) return(ENOMEM);
    sem_init(&(si -> registered), 0, 0);
    si -> start_routine = start_routine;
    si -> arg = arg;
    LOCK();
    if (!GC_thr_initialized) GC_thr_init();
#   ifdef GC_ASSERTIONS
      {
	size_t stack_size;
	if (NULL == attr) {
	   pthread_attr_t my_attr;
	   pthread_attr_init(&my_attr);
	   pthread_attr_getstacksize(&my_attr, &stack_size);
	} else {
	   pthread_attr_getstacksize(attr, &stack_size);
	}
#       ifdef PARALLEL_MARK
	  GC_ASSERT(stack_size >= (8*HBLKSIZE*sizeof(word)));
#       else
          /* FreeBSD-5.3/Alpha: default pthread stack is 64K, 	*/
	  /* HBLKSIZE=8192, sizeof(word)=8			*/
	  GC_ASSERT(stack_size >= 65536);
#       endif
	/* Our threads may need to do some work for the GC.	*/
	/* Ridiculously small threads won't work, and they	*/
	/* probably wouldn't work anyway.			*/
      }
#   endif
    if (NULL == attr) {
	detachstate = PTHREAD_CREATE_JOINABLE;
    } else { 
        pthread_attr_getdetachstate(attr, &detachstate);
    }
    if (PTHREAD_CREATE_DETACHED == detachstate) my_flags |= DETACHED;
    si -> flags = my_flags;
    UNLOCK();
#   ifdef DEBUG_THREADS
        GC_printf1("About to start new thread from thread 0x%X\n",
		   pthread_self());
#   endif

    result = REAL_FUNC(pthread_create)(new_thread, attr, GC_start_routine, si);

#   ifdef DEBUG_THREADS
        GC_printf1("Started thread 0x%X\n", *new_thread);
#   endif
    /* Wait until child has been added to the thread table.		*/
    /* This also ensures that we hold onto si until the child is done	*/
    /* with it.  Thus it doesn't matter whether it is otherwise		*/
    /* visible to the collector.					*/
    if (0 == result) {
	while (0 != sem_wait(&(si -> registered))) {
            if (EINTR != errno) ABORT("sem_wait failed");
	}
    }
    sem_destroy(&(si -> registered));
    LOCK();
    GC_INTERNAL_FREE(si);
    UNLOCK();

    return(result);
}

#ifdef GENERIC_COMPARE_AND_SWAP
  pthread_mutex_t GC_compare_and_swap_lock = PTHREAD_MUTEX_INITIALIZER;

  GC_bool GC_compare_and_exchange(volatile GC_word *addr,
  			          GC_word old, GC_word new_val)
  {
    GC_bool result;
    pthread_mutex_lock(&GC_compare_and_swap_lock);
    if (*addr == old) {
      *addr = new_val;
      result = TRUE;
    } else {
      result = FALSE;
    }
    pthread_mutex_unlock(&GC_compare_and_swap_lock);
    return result;
  }
  
  GC_word GC_atomic_add(volatile GC_word *addr, GC_word how_much)
  {
    GC_word old;
    pthread_mutex_lock(&GC_compare_and_swap_lock);
    old = *addr;
    *addr = old + how_much;
    pthread_mutex_unlock(&GC_compare_and_swap_lock);
    return old;
  }

#endif /* GENERIC_COMPARE_AND_SWAP */
/* Spend a few cycles in a way that can't introduce contention with	*/
/* othre threads.							*/
void GC_pause()
{
    int i;
#   if !defined(__GNUC__) || defined(__INTEL_COMPILER)
      volatile word dummy = 0;
#   endif

    for (i = 0; i < 10; ++i) { 
#     if defined(__GNUC__) && !defined(__INTEL_COMPILER)
        __asm__ __volatile__ (" " : : : "memory");
#     else
	/* Something that's unlikely to be optimized away. */
	GC_noop(++dummy);
#     endif
    }
}
    
#define SPIN_MAX 128	/* Maximum number of calls to GC_pause before	*/
			/* give up.					*/

VOLATILE GC_bool GC_collecting = 0;
			/* A hint that we're in the collector and       */
                        /* holding the allocation lock for an           */
                        /* extended period.                             */

#if !defined(USE_SPIN_LOCK) || defined(PARALLEL_MARK)
/* If we don't want to use the below spinlock implementation, either	*/
/* because we don't have a GC_test_and_set implementation, or because 	*/
/* we don't want to risk sleeping, we can still try spinning on 	*/
/* pthread_mutex_trylock for a while.  This appears to be very		*/
/* beneficial in many cases.						*/
/* I suspect that under high contention this is nearly always better	*/
/* than the spin lock.  But it's a bit slower on a uniprocessor.	*/
/* Hence we still default to the spin lock.				*/
/* This is also used to acquire the mark lock for the parallel		*/
/* marker.								*/

/* Here we use a strict exponential backoff scheme.  I don't know 	*/
/* whether that's better or worse than the above.  We eventually 	*/
/* yield by calling pthread_mutex_lock(); it never makes sense to	*/
/* explicitly sleep.							*/

#define LOCK_STATS
#ifdef LOCK_STATS
  unsigned long GC_spin_count = 0;
  unsigned long GC_block_count = 0;
  unsigned long GC_unlocked_count = 0;
#endif

void GC_generic_lock(pthread_mutex_t * lock)
{
#ifndef NO_PTHREAD_TRYLOCK
    unsigned pause_length = 1;
    unsigned i;
    
    if (0 == pthread_mutex_trylock(lock)) {
#       ifdef LOCK_STATS
	    ++GC_unlocked_count;
#       endif
	return;
    }
    for (; pause_length <= SPIN_MAX; pause_length <<= 1) {
	for (i = 0; i < pause_length; ++i) {
	    GC_pause();
	}
        switch(pthread_mutex_trylock(lock)) {
	    case 0:
#		ifdef LOCK_STATS
		    ++GC_spin_count;
#		endif
		return;
	    case EBUSY:
		break;
	    default:
		ABORT("Unexpected error from pthread_mutex_trylock");
        }
    }
#endif /* !NO_PTHREAD_TRYLOCK */
#   ifdef LOCK_STATS
	++GC_block_count;
#   endif
    pthread_mutex_lock(lock);
}

#endif /* !USE_SPIN_LOCK || PARALLEL_MARK */

#if defined(USE_SPIN_LOCK)

/* Reasonably fast spin locks.  Basically the same implementation */
/* as STL alloc.h.  This isn't really the right way to do this.   */
/* but until the POSIX scheduling mess gets straightened out ...  */

volatile unsigned int GC_allocate_lock = 0;


void GC_lock()
{
#   define low_spin_max 30  /* spin cycles if we suspect uniprocessor */
#   define high_spin_max SPIN_MAX /* spin cycles for multiprocessor */
    static unsigned spin_max = low_spin_max;
    unsigned my_spin_max;
    static unsigned last_spins = 0;
    unsigned my_last_spins;
    int i;

    if (!GC_test_and_set(&GC_allocate_lock)) {
        return;
    }
    my_spin_max = spin_max;
    my_last_spins = last_spins;
    for (i = 0; i < my_spin_max; i++) {
        if (GC_collecting || GC_nprocs == 1) goto yield;
        if (i < my_last_spins/2 || GC_allocate_lock) {
            GC_pause();
            continue;
        }
        if (!GC_test_and_set(&GC_allocate_lock)) {
	    /*
             * got it!
             * Spinning worked.  Thus we're probably not being scheduled
             * against the other process with which we were contending.
             * Thus it makes sense to spin longer the next time.
	     */
            last_spins = i;
            spin_max = high_spin_max;
            return;
        }
    }
    /* We are probably being scheduled against the other process.  Sleep. */
    spin_max = low_spin_max;
yield:
    for (i = 0;; ++i) {
        if (!GC_test_and_set(&GC_allocate_lock)) {
            return;
        }
#       define SLEEP_THRESHOLD 12
		/* Under Linux very short sleeps tend to wait until	*/
		/* the current time quantum expires.  On old Linux	*/
		/* kernels nanosleep(<= 2ms) just spins under Linux.    */
		/* (Under 2.4, this happens only for real-time		*/
		/* processes.)  We want to minimize both behaviors	*/
		/* here.						*/
        if (i < SLEEP_THRESHOLD) {
            sched_yield();
	} else {
	    struct timespec ts;
	
	    if (i > 24) i = 24;
			/* Don't wait for more than about 15msecs, even	*/
			/* under extreme contention.			*/
	    ts.tv_sec = 0;
	    ts.tv_nsec = 1 << i;
	    nanosleep(&ts, 0);
	}
    }
}

#else  /* !USE_SPINLOCK */
void GC_lock()
{
#ifndef NO_PTHREAD_TRYLOCK
    if (1 == GC_nprocs || GC_collecting) {
	pthread_mutex_lock(&GC_allocate_ml);
    } else {
        GC_generic_lock(&GC_allocate_ml);
    }
#else  /* !NO_PTHREAD_TRYLOCK */
    pthread_mutex_lock(&GC_allocate_ml);
#endif /* !NO_PTHREAD_TRYLOCK */
}

#endif /* !USE_SPINLOCK */

#if defined(PARALLEL_MARK) || defined(THREAD_LOCAL_ALLOC)

#ifdef GC_ASSERTIONS
  pthread_t GC_mark_lock_holder = NO_THREAD;
#endif

#if 0
  /* Ugly workaround for a linux threads bug in the final versions      */
  /* of glibc2.1.  Pthread_mutex_trylock sets the mutex owner           */
  /* field even when it fails to acquire the mutex.  This causes        */
  /* pthread_cond_wait to die.  Remove for glibc2.2.                    */
  /* According to the man page, we should use                           */
  /* PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP, but that isn't actually   */
  /* defined.                                                           */
  static pthread_mutex_t mark_mutex =
        {0, 0, 0, PTHREAD_MUTEX_ERRORCHECK_NP, {0, 0}};
#else
  static pthread_mutex_t mark_mutex = PTHREAD_MUTEX_INITIALIZER;
#endif

static pthread_cond_t builder_cv = PTHREAD_COND_INITIALIZER;

void GC_acquire_mark_lock()
{
/*
    if (pthread_mutex_lock(&mark_mutex) != 0) {
	ABORT("pthread_mutex_lock failed");
    }
*/
    GC_generic_lock(&mark_mutex);
#   ifdef GC_ASSERTIONS
	GC_mark_lock_holder = pthread_self();
#   endif
}

void GC_release_mark_lock()
{
    GC_ASSERT(GC_mark_lock_holder == pthread_self());
#   ifdef GC_ASSERTIONS
	GC_mark_lock_holder = NO_THREAD;
#   endif
    if (pthread_mutex_unlock(&mark_mutex) != 0) {
	ABORT("pthread_mutex_unlock failed");
    }
}

/* Collector must wait for a freelist builders for 2 reasons:		*/
/* 1) Mark bits may still be getting examined without lock.		*/
/* 2) Partial free lists referenced only by locals may not be scanned 	*/
/*    correctly, e.g. if they contain "pointer-free" objects, since the	*/
/*    free-list link may be ignored.					*/
void GC_wait_builder()
{
    GC_ASSERT(GC_mark_lock_holder == pthread_self());
#   ifdef GC_ASSERTIONS
	GC_mark_lock_holder = NO_THREAD;
#   endif
    if (pthread_cond_wait(&builder_cv, &mark_mutex) != 0) {
	ABORT("pthread_cond_wait failed");
    }
    GC_ASSERT(GC_mark_lock_holder == NO_THREAD);
#   ifdef GC_ASSERTIONS
	GC_mark_lock_holder = pthread_self();
#   endif
}

void GC_wait_for_reclaim()
{
    GC_acquire_mark_lock();
    while (GC_fl_builder_count > 0) {
	GC_wait_builder();
    }
    GC_release_mark_lock();
}

void GC_notify_all_builder()
{
    GC_ASSERT(GC_mark_lock_holder == pthread_self());
    if (pthread_cond_broadcast(&builder_cv) != 0) {
	ABORT("pthread_cond_broadcast failed");
    }
}

#endif /* PARALLEL_MARK || THREAD_LOCAL_ALLOC */

#ifdef PARALLEL_MARK

static pthread_cond_t mark_cv = PTHREAD_COND_INITIALIZER;

void GC_wait_marker()
{
    GC_ASSERT(GC_mark_lock_holder == pthread_self());
#   ifdef GC_ASSERTIONS
	GC_mark_lock_holder = NO_THREAD;
#   endif
    if (pthread_cond_wait(&mark_cv, &mark_mutex) != 0) {
	ABORT("pthread_cond_wait failed");
    }
    GC_ASSERT(GC_mark_lock_holder == NO_THREAD);
#   ifdef GC_ASSERTIONS
	GC_mark_lock_holder = pthread_self();
#   endif
}

void GC_notify_all_marker()
{
    if (pthread_cond_broadcast(&mark_cv) != 0) {
	ABORT("pthread_cond_broadcast failed");
    }
}

#endif /* PARALLEL_MARK */

# endif /* GC_LINUX_THREADS and friends */