summaryrefslogtreecommitdiff
path: root/libgo/runtime/proc.c
blob: 224dce95ba021b092d326680253067b737d149f6 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
// Copyright 2009 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

#include <limits.h>
#include <stdlib.h>
#include <pthread.h>
#include <unistd.h>

#include "config.h"

#ifdef HAVE_DL_ITERATE_PHDR
#include <link.h>
#endif

#include "runtime.h"
#include "arch.h"
#include "defs.h"
#include "malloc.h"
#include "go-defer.h"

#ifdef USING_SPLIT_STACK

/* FIXME: These are not declared anywhere.  */

extern void __splitstack_getcontext(void *context[10]);

extern void __splitstack_setcontext(void *context[10]);

extern void *__splitstack_makecontext(size_t, void *context[10], size_t *);

extern void * __splitstack_resetcontext(void *context[10], size_t *);

extern void *__splitstack_find(void *, void *, size_t *, void **, void **,
			       void **);

extern void __splitstack_block_signals (int *, int *);

extern void __splitstack_block_signals_context (void *context[10], int *,
						int *);

#endif

#ifndef PTHREAD_STACK_MIN
# define PTHREAD_STACK_MIN 8192
#endif

#if defined(USING_SPLIT_STACK) && defined(LINKER_SUPPORTS_SPLIT_STACK)
# define StackMin PTHREAD_STACK_MIN
#else
# define StackMin 2 * 1024 * 1024
#endif

uintptr runtime_stacks_sys;

static void schedule(G*);

static void gtraceback(G*);

typedef struct Sched Sched;

M	runtime_m0;
G	runtime_g0;	// idle goroutine for m0

#ifdef __rtems__
#define __thread
#endif

static __thread G *g;
static __thread M *m;

#ifndef SETCONTEXT_CLOBBERS_TLS

static inline void
initcontext(void)
{
}

static inline void
fixcontext(ucontext_t *c __attribute__ ((unused)))
{
}

# else

# if defined(__x86_64__) && defined(__sun__)

// x86_64 Solaris 10 and 11 have a bug: setcontext switches the %fs
// register to that of the thread which called getcontext.  The effect
// is that the address of all __thread variables changes.  This bug
// also affects pthread_self() and pthread_getspecific.  We work
// around it by clobbering the context field directly to keep %fs the
// same.

static __thread greg_t fs;

static inline void
initcontext(void)
{
	ucontext_t c;

	getcontext(&c);
	fs = c.uc_mcontext.gregs[REG_FSBASE];
}

static inline void
fixcontext(ucontext_t* c)
{
	c->uc_mcontext.gregs[REG_FSBASE] = fs;
}

# else

#  error unknown case for SETCONTEXT_CLOBBERS_TLS

# endif

#endif

// We can not always refer to the TLS variables directly.  The
// compiler will call tls_get_addr to get the address of the variable,
// and it may hold it in a register across a call to schedule.  When
// we get back from the call we may be running in a different thread,
// in which case the register now points to the TLS variable for a
// different thread.  We use non-inlinable functions to avoid this
// when necessary.

G* runtime_g(void) __attribute__ ((noinline, no_split_stack));

G*
runtime_g(void)
{
	return g;
}

M* runtime_m(void) __attribute__ ((noinline, no_split_stack));

M*
runtime_m(void)
{
	return m;
}

int32	runtime_gcwaiting;

// The static TLS size.  See runtime_newm.
static int tlssize;

#ifdef HAVE_DL_ITERATE_PHDR

// Called via dl_iterate_phdr.

static int
addtls(struct dl_phdr_info* info, size_t size __attribute__ ((unused)), void *data)
{
	size_t *total = (size_t *)data;
	unsigned int i;

	for(i = 0; i < info->dlpi_phnum; ++i) {
		if(info->dlpi_phdr[i].p_type == PT_TLS)
			*total += info->dlpi_phdr[i].p_memsz;
	}
	return 0;
}

// Set the total TLS size.

static void
inittlssize()
{
	size_t total = 0;

	dl_iterate_phdr(addtls, (void *)&total);
	tlssize = total;
}

#else

static void
inittlssize()
{
}

#endif

// Go scheduler
//
// The go scheduler's job is to match ready-to-run goroutines (`g's)
// with waiting-for-work schedulers (`m's).  If there are ready g's
// and no waiting m's, ready() will start a new m running in a new
// OS thread, so that all ready g's can run simultaneously, up to a limit.
// For now, m's never go away.
//
// By default, Go keeps only one kernel thread (m) running user code
// at a single time; other threads may be blocked in the operating system.
// Setting the environment variable $GOMAXPROCS or calling
// runtime.GOMAXPROCS() will change the number of user threads
// allowed to execute simultaneously.  $GOMAXPROCS is thus an
// approximation of the maximum number of cores to use.
//
// Even a program that can run without deadlock in a single process
// might use more m's if given the chance.  For example, the prime
// sieve will use as many m's as there are primes (up to runtime_sched.mmax),
// allowing different stages of the pipeline to execute in parallel.
// We could revisit this choice, only kicking off new m's for blocking
// system calls, but that would limit the amount of parallel computation
// that go would try to do.
//
// In general, one could imagine all sorts of refinements to the
// scheduler, but the goal now is just to get something working on
// Linux and OS X.

struct Sched {
	Lock;

	G *gfree;	// available g's (status == Gdead)
	int32 goidgen;

	G *ghead;	// g's waiting to run
	G *gtail;
	int32 gwait;	// number of g's waiting to run
	int32 gcount;	// number of g's that are alive
	int32 grunning;	// number of g's running on cpu or in syscall

	M *mhead;	// m's waiting for work
	int32 mwait;	// number of m's waiting for work
	int32 mcount;	// number of m's that have been created

	volatile uint32 atomic;	// atomic scheduling word (see below)

	int32 profilehz;	// cpu profiling rate

	bool init;  // running initialization
	bool lockmain;  // init called runtime.LockOSThread

	Note	stopped;	// one g can set waitstop and wait here for m's to stop
};

// The atomic word in sched is an atomic uint32 that
// holds these fields.
//
//	[15 bits] mcpu		number of m's executing on cpu
//	[15 bits] mcpumax	max number of m's allowed on cpu
//	[1 bit] waitstop	some g is waiting on stopped
//	[1 bit] gwaiting	gwait != 0
//
// These fields are the information needed by entersyscall
// and exitsyscall to decide whether to coordinate with the
// scheduler.  Packing them into a single machine word lets
// them use a fast path with a single atomic read/write and
// no lock/unlock.  This greatly reduces contention in
// syscall- or cgo-heavy multithreaded programs.
//
// Except for entersyscall and exitsyscall, the manipulations
// to these fields only happen while holding the schedlock,
// so the routines holding schedlock only need to worry about
// what entersyscall and exitsyscall do, not the other routines
// (which also use the schedlock).
//
// In particular, entersyscall and exitsyscall only read mcpumax,
// waitstop, and gwaiting.  They never write them.  Thus, writes to those
// fields can be done (holding schedlock) without fear of write conflicts.
// There may still be logic conflicts: for example, the set of waitstop must
// be conditioned on mcpu >= mcpumax or else the wait may be a
// spurious sleep.  The Promela model in proc.p verifies these accesses.
enum {
	mcpuWidth = 15,
	mcpuMask = (1<<mcpuWidth) - 1,
	mcpuShift = 0,
	mcpumaxShift = mcpuShift + mcpuWidth,
	waitstopShift = mcpumaxShift + mcpuWidth,
	gwaitingShift = waitstopShift+1,

	// The max value of GOMAXPROCS is constrained
	// by the max value we can store in the bit fields
	// of the atomic word.  Reserve a few high values
	// so that we can detect accidental decrement
	// beyond zero.
	maxgomaxprocs = mcpuMask - 10,
};

#define atomic_mcpu(v)		(((v)>>mcpuShift)&mcpuMask)
#define atomic_mcpumax(v)	(((v)>>mcpumaxShift)&mcpuMask)
#define atomic_waitstop(v)	(((v)>>waitstopShift)&1)
#define atomic_gwaiting(v)	(((v)>>gwaitingShift)&1)

Sched runtime_sched;
int32 runtime_gomaxprocs;
bool runtime_singleproc;

static bool canaddmcpu(void);

// An m that is waiting for notewakeup(&m->havenextg).  This may
// only be accessed while the scheduler lock is held.  This is used to
// minimize the number of times we call notewakeup while the scheduler
// lock is held, since the m will normally move quickly to lock the
// scheduler itself, producing lock contention.
static M* mwakeup;

// Scheduling helpers.  Sched must be locked.
static void gput(G*);	// put/get on ghead/gtail
static G* gget(void);
static void mput(M*);	// put/get on mhead
static M* mget(G*);
static void gfput(G*);	// put/get on gfree
static G* gfget(void);
static void matchmg(void);	// match m's to g's
static void readylocked(G*);	// ready, but sched is locked
static void mnextg(M*, G*);
static void mcommoninit(M*);

void
setmcpumax(uint32 n)
{
	uint32 v, w;

	for(;;) {
		v = runtime_sched.atomic;
		w = v;
		w &= ~(mcpuMask<<mcpumaxShift);
		w |= n<<mcpumaxShift;
		if(runtime_cas(&runtime_sched.atomic, v, w))
			break;
	}
}

// First function run by a new goroutine.  This replaces gogocall.
static void
kickoff(void)
{
	void (*fn)(void*);

	fn = (void (*)(void*))(g->entry);
	fn(g->param);
	runtime_goexit();
}

// Switch context to a different goroutine.  This is like longjmp.
static void runtime_gogo(G*) __attribute__ ((noinline));
static void
runtime_gogo(G* newg)
{
#ifdef USING_SPLIT_STACK
	__splitstack_setcontext(&newg->stack_context[0]);
#endif
	g = newg;
	newg->fromgogo = true;
	fixcontext(&newg->context);
	setcontext(&newg->context);
	runtime_throw("gogo setcontext returned");
}

// Save context and call fn passing g as a parameter.  This is like
// setjmp.  Because getcontext always returns 0, unlike setjmp, we use
// g->fromgogo as a code.  It will be true if we got here via
// setcontext.  g == nil the first time this is called in a new m.
static void runtime_mcall(void (*)(G*)) __attribute__ ((noinline));
static void
runtime_mcall(void (*pfn)(G*))
{
	M *mp;
	G *gp;
#ifndef USING_SPLIT_STACK
	int i;
#endif

	// Ensure that all registers are on the stack for the garbage
	// collector.
	__builtin_unwind_init();

	mp = m;
	gp = g;
	if(gp == mp->g0)
		runtime_throw("runtime: mcall called on m->g0 stack");

	if(gp != nil) {

#ifdef USING_SPLIT_STACK
		__splitstack_getcontext(&g->stack_context[0]);
#else
		gp->gcnext_sp = &i;
#endif
		gp->fromgogo = false;
		getcontext(&gp->context);

		// When we return from getcontext, we may be running
		// in a new thread.  That means that m and g may have
		// changed.  They are global variables so we will
		// reload them, but the addresses of m and g may be
		// cached in our local stack frame, and those
		// addresses may be wrong.  Call functions to reload
		// the values for this thread.
		mp = runtime_m();
		gp = runtime_g();

		if(gp->traceback != nil)
			gtraceback(gp);
	}
	if (gp == nil || !gp->fromgogo) {
#ifdef USING_SPLIT_STACK
		__splitstack_setcontext(&mp->g0->stack_context[0]);
#endif
		mp->g0->entry = (byte*)pfn;
		mp->g0->param = gp;

		// It's OK to set g directly here because this case
		// can not occur if we got here via a setcontext to
		// the getcontext call just above.
		g = mp->g0;

		fixcontext(&mp->g0->context);
		setcontext(&mp->g0->context);
		runtime_throw("runtime: mcall function returned");
	}
}

// Keep trace of scavenger's goroutine for deadlock detection.
static G *scvg;

// The bootstrap sequence is:
//
//	call osinit
//	call schedinit
//	make & queue new G
//	call runtime_mstart
//
// The new G calls runtime_main.
void
runtime_schedinit(void)
{
	int32 n;
	const byte *p;

	m = &runtime_m0;
	g = &runtime_g0;
	m->g0 = g;
	m->curg = g;
	g->m = m;

	initcontext();
	inittlssize();

	m->nomemprof++;
	runtime_mallocinit();
	mcommoninit(m);

	runtime_goargs();
	runtime_goenvs();

	// For debugging:
	// Allocate internal symbol table representation now,
	// so that we don't need to call malloc when we crash.
	// runtime_findfunc(0);

	runtime_gomaxprocs = 1;
	p = runtime_getenv("GOMAXPROCS");
	if(p != nil && (n = runtime_atoi(p)) != 0) {
		if(n > maxgomaxprocs)
			n = maxgomaxprocs;
		runtime_gomaxprocs = n;
	}
	// wait for the main goroutine to start before taking
	// GOMAXPROCS into account.
	setmcpumax(1);
	runtime_singleproc = runtime_gomaxprocs == 1;

	canaddmcpu();	// mcpu++ to account for bootstrap m
	m->helpgc = 1;	// flag to tell schedule() to mcpu--
	runtime_sched.grunning++;

	// Can not enable GC until all roots are registered.
	// mstats.enablegc = 1;
	m->nomemprof--;
}

extern void main_init(void) __asm__ ("__go_init_main");
extern void main_main(void) __asm__ ("main.main");

// The main goroutine.
void
runtime_main(void)
{
	// Lock the main goroutine onto this, the main OS thread,
	// during initialization.  Most programs won't care, but a few
	// do require certain calls to be made by the main thread.
	// Those can arrange for main.main to run in the main thread
	// by calling runtime.LockOSThread during initialization
	// to preserve the lock.
	runtime_LockOSThread();
	// From now on, newgoroutines may use non-main threads.
	setmcpumax(runtime_gomaxprocs);
	runtime_sched.init = true;
	scvg = __go_go(runtime_MHeap_Scavenger, nil);
	main_init();
	runtime_sched.init = false;
	if(!runtime_sched.lockmain)
		runtime_UnlockOSThread();

	// For gccgo we have to wait until after main is initialized
	// to enable GC, because initializing main registers the GC
	// roots.
	mstats.enablegc = 1;

	// The deadlock detection has false negatives.
	// Let scvg start up, to eliminate the false negative
	// for the trivial program func main() { select{} }.
	runtime_gosched();

	main_main();
	runtime_exit(0);
	for(;;)
		*(int32*)0 = 0;
}

// Lock the scheduler.
static void
schedlock(void)
{
	runtime_lock(&runtime_sched);
}

// Unlock the scheduler.
static void
schedunlock(void)
{
	M *m;

	m = mwakeup;
	mwakeup = nil;
	runtime_unlock(&runtime_sched);
	if(m != nil)
		runtime_notewakeup(&m->havenextg);
}

void
runtime_goexit(void)
{
	g->status = Gmoribund;
	runtime_gosched();
}

void
runtime_goroutineheader(G *g)
{
	const char *status;

	switch(g->status) {
	case Gidle:
		status = "idle";
		break;
	case Grunnable:
		status = "runnable";
		break;
	case Grunning:
		status = "running";
		break;
	case Gsyscall:
		status = "syscall";
		break;
	case Gwaiting:
		if(g->waitreason)
			status = g->waitreason;
		else
			status = "waiting";
		break;
	case Gmoribund:
		status = "moribund";
		break;
	default:
		status = "???";
		break;
	}
	runtime_printf("goroutine %d [%s]:\n", g->goid, status);
}

void
runtime_goroutinetrailer(G *g)
{
	if(g != nil && g->gopc != 0 && g->goid != 1) {
		struct __go_string fn;
		struct __go_string file;
		int line;

		if(__go_file_line(g->gopc - 1, &fn, &file, &line)) {
			runtime_printf("created by %s\n", fn.__data);
			runtime_printf("\t%s:%d\n", file.__data, line);
		}
	}
}

struct Traceback
{
	G* gp;
	uintptr pcbuf[100];
	int32 c;
};

void
runtime_tracebackothers(G * volatile me)
{
	G * volatile g;
	Traceback traceback;

	traceback.gp = me;
	for(g = runtime_allg; g != nil; g = g->alllink) {
		if(g == me || g->status == Gdead)
			continue;
		runtime_printf("\n");
		runtime_goroutineheader(g);

		// Our only mechanism for doing a stack trace is
		// _Unwind_Backtrace.  And that only works for the
		// current thread, not for other random goroutines.
		// So we need to switch context to the goroutine, get
		// the backtrace, and then switch back.

		// This means that if g is running or in a syscall, we
		// can't reliably print a stack trace.  FIXME.
		if(g->status == Gsyscall || g->status == Grunning) {
			runtime_printf("no stack trace available\n");
			runtime_goroutinetrailer(g);
			continue;
		}

		g->traceback = &traceback;

#ifdef USING_SPLIT_STACK
		__splitstack_getcontext(&me->stack_context[0]);
#endif
		getcontext(&me->context);

		if(g->traceback != nil) {
			runtime_gogo(g);
		}

		runtime_printtrace(traceback.pcbuf, traceback.c);
		runtime_goroutinetrailer(g);
	}
}

// Do a stack trace of gp, and then restore the context to
// gp->dotraceback.

static void
gtraceback(G* gp)
{
	Traceback* traceback;

	traceback = gp->traceback;
	gp->traceback = nil;
	traceback->c = runtime_callers(1, traceback->pcbuf,
		sizeof traceback->pcbuf / sizeof traceback->pcbuf[0]);
	runtime_gogo(traceback->gp);
}

// Mark this g as m's idle goroutine.
// This functionality might be used in environments where programs
// are limited to a single thread, to simulate a select-driven
// network server.  It is not exposed via the standard runtime API.
void
runtime_idlegoroutine(void)
{
	if(g->idlem != nil)
		runtime_throw("g is already an idle goroutine");
	g->idlem = m;
}

static void
mcommoninit(M *m)
{
	m->id = runtime_sched.mcount++;
	m->fastrand = 0x49f6428aUL + m->id + runtime_cputicks();

	if(m->mcache == nil)
		m->mcache = runtime_allocmcache();

	runtime_callers(1, m->createstack, nelem(m->createstack));

	// Add to runtime_allm so garbage collector doesn't free m
	// when it is just in a register or thread-local storage.
	m->alllink = runtime_allm;
	// runtime_NumCgoCall() iterates over allm w/o schedlock,
	// so we need to publish it safely.
	runtime_atomicstorep(&runtime_allm, m);
}

// Try to increment mcpu.  Report whether succeeded.
static bool
canaddmcpu(void)
{
	uint32 v;

	for(;;) {
		v = runtime_sched.atomic;
		if(atomic_mcpu(v) >= atomic_mcpumax(v))
			return 0;
		if(runtime_cas(&runtime_sched.atomic, v, v+(1<<mcpuShift)))
			return 1;
	}
}

// Put on `g' queue.  Sched must be locked.
static void
gput(G *g)
{
	M *m;

	// If g is wired, hand it off directly.
	if((m = g->lockedm) != nil && canaddmcpu()) {
		mnextg(m, g);
		return;
	}

	// If g is the idle goroutine for an m, hand it off.
	if(g->idlem != nil) {
		if(g->idlem->idleg != nil) {
			runtime_printf("m%d idle out of sync: g%d g%d\n",
				g->idlem->id,
				g->idlem->idleg->goid, g->goid);
			runtime_throw("runtime: double idle");
		}
		g->idlem->idleg = g;
		return;
	}

	g->schedlink = nil;
	if(runtime_sched.ghead == nil)
		runtime_sched.ghead = g;
	else
		runtime_sched.gtail->schedlink = g;
	runtime_sched.gtail = g;

	// increment gwait.
	// if it transitions to nonzero, set atomic gwaiting bit.
	if(runtime_sched.gwait++ == 0)
		runtime_xadd(&runtime_sched.atomic, 1<<gwaitingShift);
}

// Report whether gget would return something.
static bool
haveg(void)
{
	return runtime_sched.ghead != nil || m->idleg != nil;
}

// Get from `g' queue.  Sched must be locked.
static G*
gget(void)
{
	G *g;

	g = runtime_sched.ghead;
	if(g){
		runtime_sched.ghead = g->schedlink;
		if(runtime_sched.ghead == nil)
			runtime_sched.gtail = nil;
		// decrement gwait.
		// if it transitions to zero, clear atomic gwaiting bit.
		if(--runtime_sched.gwait == 0)
			runtime_xadd(&runtime_sched.atomic, -1<<gwaitingShift);
	} else if(m->idleg != nil) {
		g = m->idleg;
		m->idleg = nil;
	}
	return g;
}

// Put on `m' list.  Sched must be locked.
static void
mput(M *m)
{
	m->schedlink = runtime_sched.mhead;
	runtime_sched.mhead = m;
	runtime_sched.mwait++;
}

// Get an `m' to run `g'.  Sched must be locked.
static M*
mget(G *g)
{
	M *m;

	// if g has its own m, use it.
	if(g && (m = g->lockedm) != nil)
		return m;

	// otherwise use general m pool.
	if((m = runtime_sched.mhead) != nil){
		runtime_sched.mhead = m->schedlink;
		runtime_sched.mwait--;
	}
	return m;
}

// Mark g ready to run.
void
runtime_ready(G *g)
{
	schedlock();
	readylocked(g);
	schedunlock();
}

// Mark g ready to run.  Sched is already locked.
// G might be running already and about to stop.
// The sched lock protects g->status from changing underfoot.
static void
readylocked(G *g)
{
	if(g->m){
		// Running on another machine.
		// Ready it when it stops.
		g->readyonstop = 1;
		return;
	}

	// Mark runnable.
	if(g->status == Grunnable || g->status == Grunning) {
		runtime_printf("goroutine %d has status %d\n", g->goid, g->status);
		runtime_throw("bad g->status in ready");
	}
	g->status = Grunnable;

	gput(g);
	matchmg();
}

// Same as readylocked but a different symbol so that
// debuggers can set a breakpoint here and catch all
// new goroutines.
static void
newprocreadylocked(G *g)
{
	readylocked(g);
}

// Pass g to m for running.
// Caller has already incremented mcpu.
static void
mnextg(M *m, G *g)
{
	runtime_sched.grunning++;
	m->nextg = g;
	if(m->waitnextg) {
		m->waitnextg = 0;
		if(mwakeup != nil)
			runtime_notewakeup(&mwakeup->havenextg);
		mwakeup = m;
	}
}

// Get the next goroutine that m should run.
// Sched must be locked on entry, is unlocked on exit.
// Makes sure that at most $GOMAXPROCS g's are
// running on cpus (not in system calls) at any given time.
static G*
nextgandunlock(void)
{
	G *gp;
	uint32 v;

top:
	if(atomic_mcpu(runtime_sched.atomic) >= maxgomaxprocs)
		runtime_throw("negative mcpu");

	// If there is a g waiting as m->nextg, the mcpu++
	// happened before it was passed to mnextg.
	if(m->nextg != nil) {
		gp = m->nextg;
		m->nextg = nil;
		schedunlock();
		return gp;
	}

	if(m->lockedg != nil) {
		// We can only run one g, and it's not available.
		// Make sure some other cpu is running to handle
		// the ordinary run queue.
		if(runtime_sched.gwait != 0) {
			matchmg();
			// m->lockedg might have been on the queue.
			if(m->nextg != nil) {
				gp = m->nextg;
				m->nextg = nil;
				schedunlock();
				return gp;
			}
		}
	} else {
		// Look for work on global queue.
		while(haveg() && canaddmcpu()) {
			gp = gget();
			if(gp == nil)
				runtime_throw("gget inconsistency");

			if(gp->lockedm) {
				mnextg(gp->lockedm, gp);
				continue;
			}
			runtime_sched.grunning++;
			schedunlock();
			return gp;
		}

		// The while loop ended either because the g queue is empty
		// or because we have maxed out our m procs running go
		// code (mcpu >= mcpumax).  We need to check that
		// concurrent actions by entersyscall/exitsyscall cannot
		// invalidate the decision to end the loop.
		//
		// We hold the sched lock, so no one else is manipulating the
		// g queue or changing mcpumax.  Entersyscall can decrement
		// mcpu, but if does so when there is something on the g queue,
		// the gwait bit will be set, so entersyscall will take the slow path
		// and use the sched lock.  So it cannot invalidate our decision.
		//
		// Wait on global m queue.
		mput(m);
	}

	// Look for deadlock situation.
	// There is a race with the scavenger that causes false negatives:
	// if the scavenger is just starting, then we have
	//	scvg != nil && grunning == 0 && gwait == 0
	// and we do not detect a deadlock.  It is possible that we should
	// add that case to the if statement here, but it is too close to Go 1
	// to make such a subtle change.  Instead, we work around the
	// false negative in trivial programs by calling runtime.gosched
	// from the main goroutine just before main.main.
	// See runtime_main above.
	//
	// On a related note, it is also possible that the scvg == nil case is
	// wrong and should include gwait, but that does not happen in
	// standard Go programs, which all start the scavenger.
	//
	if((scvg == nil && runtime_sched.grunning == 0) ||
	   (scvg != nil && runtime_sched.grunning == 1 && runtime_sched.gwait == 0 &&
	    (scvg->status == Grunning || scvg->status == Gsyscall))) {
		runtime_throw("all goroutines are asleep - deadlock!");
	}

	m->nextg = nil;
	m->waitnextg = 1;
	runtime_noteclear(&m->havenextg);

	// Stoptheworld is waiting for all but its cpu to go to stop.
	// Entersyscall might have decremented mcpu too, but if so
	// it will see the waitstop and take the slow path.
	// Exitsyscall never increments mcpu beyond mcpumax.
	v = runtime_atomicload(&runtime_sched.atomic);
	if(atomic_waitstop(v) && atomic_mcpu(v) <= atomic_mcpumax(v)) {
		// set waitstop = 0 (known to be 1)
		runtime_xadd(&runtime_sched.atomic, -1<<waitstopShift);
		runtime_notewakeup(&runtime_sched.stopped);
	}
	schedunlock();

	runtime_notesleep(&m->havenextg);
	if(m->helpgc) {
		runtime_gchelper();
		m->helpgc = 0;
		runtime_lock(&runtime_sched);
		goto top;
	}
	if((gp = m->nextg) == nil)
		runtime_throw("bad m->nextg in nextgoroutine");
	m->nextg = nil;
	return gp;
}

int32
runtime_helpgc(bool *extra)
{
	M *mp;
	int32 n, max;

	// Figure out how many CPUs to use.
	// Limited by gomaxprocs, number of actual CPUs, and MaxGcproc.
	max = runtime_gomaxprocs;
	if(max > runtime_ncpu)
		max = runtime_ncpu > 0 ? runtime_ncpu : 1;
	if(max > MaxGcproc)
		max = MaxGcproc;

	// We're going to use one CPU no matter what.
	// Figure out the max number of additional CPUs.
	max--;

	runtime_lock(&runtime_sched);
	n = 0;
	while(n < max && (mp = mget(nil)) != nil) {
		n++;
		mp->helpgc = 1;
		mp->waitnextg = 0;
		runtime_notewakeup(&mp->havenextg);
	}
	runtime_unlock(&runtime_sched);
	if(extra)
		*extra = n != max;
	return n;
}

void
runtime_stoptheworld(void)
{
	uint32 v;

	schedlock();
	runtime_gcwaiting = 1;

	setmcpumax(1);

	// while mcpu > 1
	for(;;) {
		v = runtime_sched.atomic;
		if(atomic_mcpu(v) <= 1)
			break;

		// It would be unsafe for multiple threads to be using
		// the stopped note at once, but there is only
		// ever one thread doing garbage collection.
		runtime_noteclear(&runtime_sched.stopped);
		if(atomic_waitstop(v))
			runtime_throw("invalid waitstop");

		// atomic { waitstop = 1 }, predicated on mcpu <= 1 check above
		// still being true.
		if(!runtime_cas(&runtime_sched.atomic, v, v+(1<<waitstopShift)))
			continue;

		schedunlock();
		runtime_notesleep(&runtime_sched.stopped);
		schedlock();
	}
	runtime_singleproc = runtime_gomaxprocs == 1;
	schedunlock();
}

void
runtime_starttheworld(bool extra)
{
	M *m;

	schedlock();
	runtime_gcwaiting = 0;
	setmcpumax(runtime_gomaxprocs);
	matchmg();
	if(extra && canaddmcpu()) {
		// Start a new m that will (we hope) be idle
		// and so available to help when the next
		// garbage collection happens.
		// canaddmcpu above did mcpu++
		// (necessary, because m will be doing various
		// initialization work so is definitely running),
		// but m is not running a specific goroutine,
		// so set the helpgc flag as a signal to m's
		// first schedule(nil) to mcpu-- and grunning--.
		m = runtime_newm();
		m->helpgc = 1;
		runtime_sched.grunning++;
	}
	schedunlock();
}

// Called to start an M.
void*
runtime_mstart(void* mp)
{
	m = (M*)mp;
	g = m->g0;

	initcontext();

	g->entry = nil;
	g->param = nil;

	// Record top of stack for use by mcall.
	// Once we call schedule we're never coming back,
	// so other calls can reuse this stack space.
#ifdef USING_SPLIT_STACK
	__splitstack_getcontext(&g->stack_context[0]);
#else
	g->gcinitial_sp = &mp;
	// Setting gcstack_size to 0 is a marker meaning that gcinitial_sp
	// is the top of the stack, not the bottom.
	g->gcstack_size = 0;
	g->gcnext_sp = &mp;
#endif
	getcontext(&g->context);

	if(g->entry != nil) {
		// Got here from mcall.
		void (*pfn)(G*) = (void (*)(G*))g->entry;
		G* gp = (G*)g->param;
		pfn(gp);
		*(int*)0x21 = 0x21;
	}
	runtime_minit();

#ifdef USING_SPLIT_STACK
	{
	  int dont_block_signals = 0;
	  __splitstack_block_signals(&dont_block_signals, nil);
	}
#endif

	// Install signal handlers; after minit so that minit can
	// prepare the thread to be able to handle the signals.
	if(m == &runtime_m0)
		runtime_initsig();

	schedule(nil);
	return nil;
}

typedef struct CgoThreadStart CgoThreadStart;
struct CgoThreadStart
{
	M *m;
	G *g;
	void (*fn)(void);
};

// Kick off new m's as needed (up to mcpumax).
// Sched is locked.
static void
matchmg(void)
{
	G *gp;
	M *mp;

	if(m->mallocing || m->gcing)
		return;

	while(haveg() && canaddmcpu()) {
		gp = gget();
		if(gp == nil)
			runtime_throw("gget inconsistency");

		// Find the m that will run gp.
		if((mp = mget(gp)) == nil)
			mp = runtime_newm();
		mnextg(mp, gp);
	}
}

// Create a new m.  It will start off with a call to runtime_mstart.
M*
runtime_newm(void)
{
	M *m;
	pthread_attr_t attr;
	pthread_t tid;
	size_t stacksize;

	m = runtime_malloc(sizeof(M));
	mcommoninit(m);
	m->g0 = runtime_malg(-1, nil, nil);

	if(pthread_attr_init(&attr) != 0)
		runtime_throw("pthread_attr_init");
	if(pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED) != 0)
		runtime_throw("pthread_attr_setdetachstate");

	stacksize = PTHREAD_STACK_MIN;

	// With glibc before version 2.16 the static TLS size is taken
	// out of the stack size, and we get an error or a crash if
	// there is not enough stack space left.  Add it back in if we
	// can, in case the program uses a lot of TLS space.  FIXME:
	// This can be disabled in glibc 2.16 and later, if the bug is
	// indeed fixed then.
	stacksize += tlssize;

	if(pthread_attr_setstacksize(&attr, stacksize) != 0)
		runtime_throw("pthread_attr_setstacksize");

	if(pthread_create(&tid, &attr, runtime_mstart, m) != 0)
		runtime_throw("pthread_create");

	return m;
}

// One round of scheduler: find a goroutine and run it.
// The argument is the goroutine that was running before
// schedule was called, or nil if this is the first call.
// Never returns.
static void
schedule(G *gp)
{
	int32 hz;
	uint32 v;

	schedlock();
	if(gp != nil) {
		// Just finished running gp.
		gp->m = nil;
		runtime_sched.grunning--;

		// atomic { mcpu-- }
		v = runtime_xadd(&runtime_sched.atomic, -1<<mcpuShift);
		if(atomic_mcpu(v) > maxgomaxprocs)
			runtime_throw("negative mcpu in scheduler");

		switch(gp->status){
		case Grunnable:
		case Gdead:
			// Shouldn't have been running!
			runtime_throw("bad gp->status in sched");
		case Grunning:
			gp->status = Grunnable;
			gput(gp);
			break;
		case Gmoribund:
			gp->status = Gdead;
			if(gp->lockedm) {
				gp->lockedm = nil;
				m->lockedg = nil;
			}
			gp->idlem = nil;
			runtime_memclr(&gp->context, sizeof gp->context);
			gfput(gp);
			if(--runtime_sched.gcount == 0)
				runtime_exit(0);
			break;
		}
		if(gp->readyonstop){
			gp->readyonstop = 0;
			readylocked(gp);
		}
	} else if(m->helpgc) {
		// Bootstrap m or new m started by starttheworld.
		// atomic { mcpu-- }
		v = runtime_xadd(&runtime_sched.atomic, -1<<mcpuShift);
		if(atomic_mcpu(v) > maxgomaxprocs)
			runtime_throw("negative mcpu in scheduler");
		// Compensate for increment in starttheworld().
		runtime_sched.grunning--;
		m->helpgc = 0;
	} else if(m->nextg != nil) {
		// New m started by matchmg.
	} else {
		runtime_throw("invalid m state in scheduler");
	}

	// Find (or wait for) g to run.  Unlocks runtime_sched.
	gp = nextgandunlock();
	gp->readyonstop = 0;
	gp->status = Grunning;
	m->curg = gp;
	gp->m = m;

	// Check whether the profiler needs to be turned on or off.
	hz = runtime_sched.profilehz;
	if(m->profilehz != hz)
		runtime_resetcpuprofiler(hz);

	runtime_gogo(gp);
}

// Enter scheduler.  If g->status is Grunning,
// re-queues g and runs everyone else who is waiting
// before running g again.  If g->status is Gmoribund,
// kills off g.
void
runtime_gosched(void)
{
	if(m->locks != 0)
		runtime_throw("gosched holding locks");
	if(g == m->g0)
		runtime_throw("gosched of g0");
	runtime_mcall(schedule);
}

// The goroutine g is about to enter a system call.
// Record that it's not using the cpu anymore.
// This is called only from the go syscall library and cgocall,
// not from the low-level system calls used by the runtime.
//
// Entersyscall cannot split the stack: the runtime_gosave must
// make g->sched refer to the caller's stack segment, because
// entersyscall is going to return immediately after.
// It's okay to call matchmg and notewakeup even after
// decrementing mcpu, because we haven't released the
// sched lock yet, so the garbage collector cannot be running.

void runtime_entersyscall(void) __attribute__ ((no_split_stack));

void
runtime_entersyscall(void)
{
	uint32 v;

	if(m->profilehz > 0)
		runtime_setprof(false);

	// Leave SP around for gc and traceback.
#ifdef USING_SPLIT_STACK
	g->gcstack = __splitstack_find(nil, nil, &g->gcstack_size,
				       &g->gcnext_segment, &g->gcnext_sp,
				       &g->gcinitial_sp);
#else
	g->gcnext_sp = (byte *) &v;
#endif

	// Save the registers in the g structure so that any pointers
	// held in registers will be seen by the garbage collector.
	getcontext(&g->gcregs);

	g->status = Gsyscall;

	// Fast path.
	// The slow path inside the schedlock/schedunlock will get
	// through without stopping if it does:
	//	mcpu--
	//	gwait not true
	//	waitstop && mcpu <= mcpumax not true
	// If we can do the same with a single atomic add,
	// then we can skip the locks.
	v = runtime_xadd(&runtime_sched.atomic, -1<<mcpuShift);
	if(!atomic_gwaiting(v) && (!atomic_waitstop(v) || atomic_mcpu(v) > atomic_mcpumax(v)))
		return;

	schedlock();
	v = runtime_atomicload(&runtime_sched.atomic);
	if(atomic_gwaiting(v)) {
		matchmg();
		v = runtime_atomicload(&runtime_sched.atomic);
	}
	if(atomic_waitstop(v) && atomic_mcpu(v) <= atomic_mcpumax(v)) {
		runtime_xadd(&runtime_sched.atomic, -1<<waitstopShift);
		runtime_notewakeup(&runtime_sched.stopped);
	}

	schedunlock();
}

// The goroutine g exited its system call.
// Arrange for it to run on a cpu again.
// This is called only from the go syscall library, not
// from the low-level system calls used by the runtime.
void
runtime_exitsyscall(void)
{
	G *gp;
	uint32 v;

	// Fast path.
	// If we can do the mcpu++ bookkeeping and
	// find that we still have mcpu <= mcpumax, then we can
	// start executing Go code immediately, without having to
	// schedlock/schedunlock.
	// Also do fast return if any locks are held, so that
	// panic code can use syscalls to open a file.
	gp = g;
	v = runtime_xadd(&runtime_sched.atomic, (1<<mcpuShift));
	if((m->profilehz == runtime_sched.profilehz && atomic_mcpu(v) <= atomic_mcpumax(v)) || m->locks > 0) {
		// There's a cpu for us, so we can run.
		gp->status = Grunning;
		// Garbage collector isn't running (since we are),
		// so okay to clear gcstack.
#ifdef USING_SPLIT_STACK
		gp->gcstack = nil;
#endif
		gp->gcnext_sp = nil;
		runtime_memclr(&gp->gcregs, sizeof gp->gcregs);

		if(m->profilehz > 0)
			runtime_setprof(true);
		return;
	}

	// Tell scheduler to put g back on the run queue:
	// mostly equivalent to g->status = Grunning,
	// but keeps the garbage collector from thinking
	// that g is running right now, which it's not.
	gp->readyonstop = 1;

	// All the cpus are taken.
	// The scheduler will ready g and put this m to sleep.
	// When the scheduler takes g away from m,
	// it will undo the runtime_sched.mcpu++ above.
	runtime_gosched();

	// Gosched returned, so we're allowed to run now.
	// Delete the gcstack information that we left for
	// the garbage collector during the system call.
	// Must wait until now because until gosched returns
	// we don't know for sure that the garbage collector
	// is not running.
#ifdef USING_SPLIT_STACK
	gp->gcstack = nil;
#endif
	gp->gcnext_sp = nil;
	runtime_memclr(&gp->gcregs, sizeof gp->gcregs);
}

// Allocate a new g, with a stack big enough for stacksize bytes.
G*
runtime_malg(int32 stacksize, byte** ret_stack, size_t* ret_stacksize)
{
	G *newg;

	newg = runtime_malloc(sizeof(G));
	if(stacksize >= 0) {
#if USING_SPLIT_STACK
		int dont_block_signals = 0;

		*ret_stack = __splitstack_makecontext(stacksize,
						      &newg->stack_context[0],
						      ret_stacksize);
		__splitstack_block_signals_context(&newg->stack_context[0],
						   &dont_block_signals, nil);
#else
		*ret_stack = runtime_mallocgc(stacksize, FlagNoProfiling|FlagNoGC, 0, 0);
		*ret_stacksize = stacksize;
		newg->gcinitial_sp = *ret_stack;
		newg->gcstack_size = stacksize;
		runtime_xadd(&runtime_stacks_sys, stacksize);
#endif
	}
	return newg;
}

/* For runtime package testing.  */

void runtime_testing_entersyscall(void)
  __asm__("runtime.entersyscall");

void
runtime_testing_entersyscall()
{
	runtime_entersyscall();
}

void runtime_testing_exitsyscall(void)
  __asm__("runtime.exitsyscall");

void
runtime_testing_exitsyscall()
{
	runtime_exitsyscall();
}

G*
__go_go(void (*fn)(void*), void* arg)
{
	byte *sp;
	size_t spsize;
	G *newg;

	schedlock();

	if((newg = gfget()) != nil){
#ifdef USING_SPLIT_STACK
		int dont_block_signals = 0;

		sp = __splitstack_resetcontext(&newg->stack_context[0],
					       &spsize);
		__splitstack_block_signals_context(&newg->stack_context[0],
						   &dont_block_signals, nil);
#else
		sp = newg->gcinitial_sp;
		spsize = newg->gcstack_size;
		if(spsize == 0)
			runtime_throw("bad spsize in __go_go");
		newg->gcnext_sp = sp;
#endif
	} else {
		newg = runtime_malg(StackMin, &sp, &spsize);
		if(runtime_lastg == nil)
			runtime_allg = newg;
		else
			runtime_lastg->alllink = newg;
		runtime_lastg = newg;
	}
	newg->status = Gwaiting;
	newg->waitreason = "new goroutine";

	newg->entry = (byte*)fn;
	newg->param = arg;
	newg->gopc = (uintptr)__builtin_return_address(0);

	runtime_sched.gcount++;
	runtime_sched.goidgen++;
	newg->goid = runtime_sched.goidgen;

	if(sp == nil)
		runtime_throw("nil g->stack0");

	{
		// Avoid warnings about variables clobbered by
		// longjmp.
		byte * volatile vsp = sp;
		size_t volatile vspsize = spsize;
		G * volatile vnewg = newg;

		getcontext(&vnewg->context);
		vnewg->context.uc_stack.ss_sp = vsp;
#ifdef MAKECONTEXT_STACK_TOP
		vnewg->context.uc_stack.ss_sp += vspsize;
#endif
		vnewg->context.uc_stack.ss_size = vspsize;
		makecontext(&vnewg->context, kickoff, 0);

		newprocreadylocked(vnewg);
		schedunlock();

		return vnewg;
	}
}

// Put on gfree list.  Sched must be locked.
static void
gfput(G *g)
{
	g->schedlink = runtime_sched.gfree;
	runtime_sched.gfree = g;
}

// Get from gfree list.  Sched must be locked.
static G*
gfget(void)
{
	G *g;

	g = runtime_sched.gfree;
	if(g)
		runtime_sched.gfree = g->schedlink;
	return g;
}

// Run all deferred functions for the current goroutine.
static void
rundefer(void)
{
	Defer *d;

	while((d = g->defer) != nil) {
		void (*pfn)(void*);

		pfn = d->__pfn;
		d->__pfn = nil;
		if (pfn != nil)
			(*pfn)(d->__arg);
		g->defer = d->__next;
		runtime_free(d);
	}
}

void runtime_Goexit (void) asm ("runtime.Goexit");

void
runtime_Goexit(void)
{
	rundefer();
	runtime_goexit();
}

void runtime_Gosched (void) asm ("runtime.Gosched");

void
runtime_Gosched(void)
{
	runtime_gosched();
}

// Implementation of runtime.GOMAXPROCS.
// delete when scheduler is stronger
int32
runtime_gomaxprocsfunc(int32 n)
{
	int32 ret;
	uint32 v;

	schedlock();
	ret = runtime_gomaxprocs;
	if(n <= 0)
		n = ret;
	if(n > maxgomaxprocs)
		n = maxgomaxprocs;
	runtime_gomaxprocs = n;
	if(runtime_gomaxprocs > 1)
		runtime_singleproc = false;
 	if(runtime_gcwaiting != 0) {
 		if(atomic_mcpumax(runtime_sched.atomic) != 1)
 			runtime_throw("invalid mcpumax during gc");
		schedunlock();
		return ret;
	}

	setmcpumax(n);

	// If there are now fewer allowed procs
	// than procs running, stop.
	v = runtime_atomicload(&runtime_sched.atomic);
	if((int32)atomic_mcpu(v) > n) {
		schedunlock();
		runtime_gosched();
		return ret;
	}
	// handle more procs
	matchmg();
	schedunlock();
	return ret;
}

void
runtime_LockOSThread(void)
{
	if(m == &runtime_m0 && runtime_sched.init) {
		runtime_sched.lockmain = true;
		return;
	}
	m->lockedg = g;
	g->lockedm = m;
}

void
runtime_UnlockOSThread(void)
{
	if(m == &runtime_m0 && runtime_sched.init) {
		runtime_sched.lockmain = false;
		return;
	}
	m->lockedg = nil;
	g->lockedm = nil;
}

bool
runtime_lockedOSThread(void)
{
	return g->lockedm != nil && m->lockedg != nil;
}

// for testing of callbacks

_Bool runtime_golockedOSThread(void)
  asm("runtime.golockedOSThread");

_Bool
runtime_golockedOSThread(void)
{
	return runtime_lockedOSThread();
}

// for testing of wire, unwire
uint32
runtime_mid()
{
	return m->id;
}

int32 runtime_NumGoroutine (void)
  __asm__ ("runtime.NumGoroutine");

int32
runtime_NumGoroutine()
{
	return runtime_sched.gcount;
}

int32
runtime_gcount(void)
{
	return runtime_sched.gcount;
}

int32
runtime_mcount(void)
{
	return runtime_sched.mcount;
}

static struct {
	Lock;
	void (*fn)(uintptr*, int32);
	int32 hz;
	uintptr pcbuf[100];
} prof;

// Called if we receive a SIGPROF signal.
void
runtime_sigprof()
{
	int32 n;

	if(prof.fn == nil || prof.hz == 0)
		return;

	runtime_lock(&prof);
	if(prof.fn == nil) {
		runtime_unlock(&prof);
		return;
	}
	n = runtime_callers(0, prof.pcbuf, nelem(prof.pcbuf));
	if(n > 0)
		prof.fn(prof.pcbuf, n);
	runtime_unlock(&prof);
}

// Arrange to call fn with a traceback hz times a second.
void
runtime_setcpuprofilerate(void (*fn)(uintptr*, int32), int32 hz)
{
	// Force sane arguments.
	if(hz < 0)
		hz = 0;
	if(hz == 0)
		fn = nil;
	if(fn == nil)
		hz = 0;

	// Stop profiler on this cpu so that it is safe to lock prof.
	// if a profiling signal came in while we had prof locked,
	// it would deadlock.
	runtime_resetcpuprofiler(0);

	runtime_lock(&prof);
	prof.fn = fn;
	prof.hz = hz;
	runtime_unlock(&prof);
	runtime_lock(&runtime_sched);
	runtime_sched.profilehz = hz;
	runtime_unlock(&runtime_sched);

	if(hz != 0)
		runtime_resetcpuprofiler(hz);
}