aboutsummaryrefslogtreecommitdiff
path: root/sys/kern/subr_smp.c
blob: 935fb6ee977cdbe5c2a93a60fdea520fbf194304 (plain) (blame)
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
/*-
 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
 *
 * Copyright (c) 2001, John Baldwin <jhb@FreeBSD.org>.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

/*
 * This module holds the global variables and machine independent functions
 * used for the kernel SMP support.
 */

#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/ktr.h>
#include <sys/proc.h>
#include <sys/bus.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mutex.h>
#include <sys/pcpu.h>
#include <sys/sched.h>
#include <sys/smp.h>
#include <sys/sysctl.h>

#include <machine/cpu.h>
#include <machine/smp.h>

#include "opt_sched.h"

#ifdef SMP
MALLOC_DEFINE(M_TOPO, "toponodes", "SMP topology data");

volatile cpuset_t stopped_cpus;
volatile cpuset_t started_cpus;
volatile cpuset_t suspended_cpus;
cpuset_t hlt_cpus_mask;
cpuset_t logical_cpus_mask;

void (*cpustop_restartfunc)(void);
#endif

static int sysctl_kern_smp_active(SYSCTL_HANDLER_ARGS);

/* This is used in modules that need to work in both SMP and UP. */
cpuset_t all_cpus;

int mp_ncpus;
/* export this for libkvm consumers. */
int mp_maxcpus = MAXCPU;

volatile int smp_started;
u_int mp_maxid;

static SYSCTL_NODE(_kern, OID_AUTO, smp,
    CTLFLAG_RD | CTLFLAG_CAPRD | CTLFLAG_MPSAFE, NULL,
    "Kernel SMP");

SYSCTL_INT(_kern_smp, OID_AUTO, maxid, CTLFLAG_RD|CTLFLAG_CAPRD, &mp_maxid, 0,
    "Max CPU ID.");

SYSCTL_INT(_kern_smp, OID_AUTO, maxcpus, CTLFLAG_RD|CTLFLAG_CAPRD, &mp_maxcpus,
    0, "Max number of CPUs that the system was compiled for.");

SYSCTL_PROC(_kern_smp, OID_AUTO, active, CTLFLAG_RD|CTLTYPE_INT|CTLFLAG_MPSAFE,
    NULL, 0, sysctl_kern_smp_active, "I",
    "Indicates system is running in SMP mode");

int smp_disabled = 0;	/* has smp been disabled? */
SYSCTL_INT(_kern_smp, OID_AUTO, disabled, CTLFLAG_RDTUN|CTLFLAG_CAPRD,
    &smp_disabled, 0, "SMP has been disabled from the loader");

int smp_cpus = 1;	/* how many cpu's running */
SYSCTL_INT(_kern_smp, OID_AUTO, cpus, CTLFLAG_RD|CTLFLAG_CAPRD, &smp_cpus, 0,
    "Number of CPUs online");

int smp_threads_per_core = 1;	/* how many SMT threads are running per core */
SYSCTL_INT(_kern_smp, OID_AUTO, threads_per_core, CTLFLAG_RD|CTLFLAG_CAPRD,
    &smp_threads_per_core, 0, "Number of SMT threads online per core");

int mp_ncores = -1;	/* how many physical cores running */
SYSCTL_INT(_kern_smp, OID_AUTO, cores, CTLFLAG_RD|CTLFLAG_CAPRD, &mp_ncores, 0,
    "Number of physical cores online");

int smp_topology = 0;	/* Which topology we're using. */
SYSCTL_INT(_kern_smp, OID_AUTO, topology, CTLFLAG_RDTUN, &smp_topology, 0,
    "Topology override setting; 0 is default provided by hardware.");

#ifdef SMP
/* Enable forwarding of a signal to a process running on a different CPU */
static int forward_signal_enabled = 1;
SYSCTL_INT(_kern_smp, OID_AUTO, forward_signal_enabled, CTLFLAG_RW,
	   &forward_signal_enabled, 0,
	   "Forwarding of a signal to a process on a different CPU");

/* Variables needed for SMP rendezvous. */
static volatile int smp_rv_ncpus;
static void (*volatile smp_rv_setup_func)(void *arg);
static void (*volatile smp_rv_action_func)(void *arg);
static void (*volatile smp_rv_teardown_func)(void *arg);
static void *volatile smp_rv_func_arg;
static volatile int smp_rv_waiters[4];

/* 
 * Shared mutex to restrict busywaits between smp_rendezvous() and
 * smp(_targeted)_tlb_shootdown().  A deadlock occurs if both of these
 * functions trigger at once and cause multiple CPUs to busywait with
 * interrupts disabled. 
 */
struct mtx smp_ipi_mtx;

/*
 * Let the MD SMP code initialize mp_maxid very early if it can.
 */
static void
mp_setmaxid(void *dummy)
{

	cpu_mp_setmaxid();

	KASSERT(mp_ncpus >= 1, ("%s: CPU count < 1", __func__));
	KASSERT(mp_ncpus > 1 || mp_maxid == 0,
	    ("%s: one CPU but mp_maxid is not zero", __func__));
	KASSERT(mp_maxid >= mp_ncpus - 1,
	    ("%s: counters out of sync: max %d, count %d", __func__,
		mp_maxid, mp_ncpus));
}
SYSINIT(cpu_mp_setmaxid, SI_SUB_TUNABLES, SI_ORDER_FIRST, mp_setmaxid, NULL);

/*
 * Call the MD SMP initialization code.
 */
static void
mp_start(void *dummy)
{

	mtx_init(&smp_ipi_mtx, "smp rendezvous", NULL, MTX_SPIN);

	/* Probe for MP hardware. */
	if (smp_disabled != 0 || cpu_mp_probe() == 0) {
		mp_ncores = 1;
		mp_ncpus = 1;
		CPU_SETOF(PCPU_GET(cpuid), &all_cpus);
		return;
	}

	cpu_mp_start();
	printf("FreeBSD/SMP: Multiprocessor System Detected: %d CPUs\n",
	    mp_ncpus);

	/* Provide a default for most architectures that don't have SMT/HTT. */
	if (mp_ncores < 0)
		mp_ncores = mp_ncpus;

	cpu_mp_announce();
}
SYSINIT(cpu_mp, SI_SUB_CPU, SI_ORDER_THIRD, mp_start, NULL);

void
forward_signal(struct thread *td)
{
	int id;

	/*
	 * signotify() has already set TDF_ASTPENDING and TDF_NEEDSIGCHECK on
	 * this thread, so all we need to do is poke it if it is currently
	 * executing so that it executes ast().
	 */
	THREAD_LOCK_ASSERT(td, MA_OWNED);
	KASSERT(TD_IS_RUNNING(td),
	    ("forward_signal: thread is not TDS_RUNNING"));

	CTR1(KTR_SMP, "forward_signal(%p)", td->td_proc);

	if (!smp_started || cold || KERNEL_PANICKED())
		return;
	if (!forward_signal_enabled)
		return;

	/* No need to IPI ourself. */
	if (td == curthread)
		return;

	id = td->td_oncpu;
	if (id == NOCPU)
		return;
	ipi_cpu(id, IPI_AST);
}

/*
 * When called the executing CPU will send an IPI to all other CPUs
 *  requesting that they halt execution.
 *
 * Usually (but not necessarily) called with 'other_cpus' as its arg.
 *
 *  - Signals all CPUs in map to stop.
 *  - Waits for each to stop.
 *
 * Returns:
 *  -1: error
 *   0: NA
 *   1: ok
 *
 */
#if defined(__amd64__) || defined(__i386__)
#define	X86	1
#else
#define	X86	0
#endif
static int
generic_stop_cpus(cpuset_t map, u_int type)
{
#ifdef KTR
	char cpusetbuf[CPUSETBUFSIZ];
#endif
	static volatile u_int stopping_cpu = NOCPU;
	int i;
	volatile cpuset_t *cpus;

	KASSERT(
	    type == IPI_STOP || type == IPI_STOP_HARD
#if X86
	    || type == IPI_SUSPEND
#endif
	    , ("%s: invalid stop type", __func__));

	if (!smp_started)
		return (0);

	CTR2(KTR_SMP, "stop_cpus(%s) with %u type",
	    cpusetobj_strprint(cpusetbuf, &map), type);

#if X86
	/*
	 * When suspending, ensure there are are no IPIs in progress.
	 * IPIs that have been issued, but not yet delivered (e.g.
	 * not pending on a vCPU when running under virtualization)
	 * will be lost, violating FreeBSD's assumption of reliable
	 * IPI delivery.
	 */
	if (type == IPI_SUSPEND)
		mtx_lock_spin(&smp_ipi_mtx);
#endif

#if X86
	if (!nmi_is_broadcast || nmi_kdb_lock == 0) {
#endif
	if (stopping_cpu != PCPU_GET(cpuid))
		while (atomic_cmpset_int(&stopping_cpu, NOCPU,
		    PCPU_GET(cpuid)) == 0)
			while (stopping_cpu != NOCPU)
				cpu_spinwait(); /* spin */

	/* send the stop IPI to all CPUs in map */
	ipi_selected(map, type);
#if X86
	}
#endif

#if X86
	if (type == IPI_SUSPEND)
		cpus = &suspended_cpus;
	else
#endif
		cpus = &stopped_cpus;

	i = 0;
	while (!CPU_SUBSET(cpus, &map)) {
		/* spin */
		cpu_spinwait();
		i++;
		if (i == 100000000) {
			printf("timeout stopping cpus\n");
			break;
		}
	}

#if X86
	if (type == IPI_SUSPEND)
		mtx_unlock_spin(&smp_ipi_mtx);
#endif

	stopping_cpu = NOCPU;
	return (1);
}

int
stop_cpus(cpuset_t map)
{

	return (generic_stop_cpus(map, IPI_STOP));
}

int
stop_cpus_hard(cpuset_t map)
{

	return (generic_stop_cpus(map, IPI_STOP_HARD));
}

#if X86
int
suspend_cpus(cpuset_t map)
{

	return (generic_stop_cpus(map, IPI_SUSPEND));
}
#endif

/*
 * Called by a CPU to restart stopped CPUs. 
 *
 * Usually (but not necessarily) called with 'stopped_cpus' as its arg.
 *
 *  - Signals all CPUs in map to restart.
 *  - Waits for each to restart.
 *
 * Returns:
 *  -1: error
 *   0: NA
 *   1: ok
 */
static int
generic_restart_cpus(cpuset_t map, u_int type)
{
#ifdef KTR
	char cpusetbuf[CPUSETBUFSIZ];
#endif
	volatile cpuset_t *cpus;

#if X86
	KASSERT(type == IPI_STOP || type == IPI_STOP_HARD
	    || type == IPI_SUSPEND, ("%s: invalid stop type", __func__));

	if (!smp_started)
		return (0);

	CTR1(KTR_SMP, "restart_cpus(%s)", cpusetobj_strprint(cpusetbuf, &map));

	if (type == IPI_SUSPEND)
		cpus = &resuming_cpus;
	else
		cpus = &stopped_cpus;

	/* signal other cpus to restart */
	if (type == IPI_SUSPEND)
		CPU_COPY_STORE_REL(&map, &toresume_cpus);
	else
		CPU_COPY_STORE_REL(&map, &started_cpus);

	/*
	 * Wake up any CPUs stopped with MWAIT.  From MI code we can't tell if
	 * MONITOR/MWAIT is enabled, but the potentially redundant writes are
	 * relatively inexpensive.
	 */
	if (type == IPI_STOP) {
		struct monitorbuf *mb;
		u_int id;

		CPU_FOREACH(id) {
			if (!CPU_ISSET(id, &map))
				continue;

			mb = &pcpu_find(id)->pc_monitorbuf;
			atomic_store_int(&mb->stop_state,
			    MONITOR_STOPSTATE_RUNNING);
		}
	}

	if (!nmi_is_broadcast || nmi_kdb_lock == 0) {
		/* wait for each to clear its bit */
		while (CPU_OVERLAP(cpus, &map))
			cpu_spinwait();
	}
#else /* !X86 */
	KASSERT(type == IPI_STOP || type == IPI_STOP_HARD,
	    ("%s: invalid stop type", __func__));

	if (!smp_started)
		return (0);

	CTR1(KTR_SMP, "restart_cpus(%s)", cpusetobj_strprint(cpusetbuf, &map));

	cpus = &stopped_cpus;

	/* signal other cpus to restart */
	CPU_COPY_STORE_REL(&map, &started_cpus);

	/* wait for each to clear its bit */
	while (CPU_OVERLAP(cpus, &map))
		cpu_spinwait();
#endif
	return (1);
}

int
restart_cpus(cpuset_t map)
{

	return (generic_restart_cpus(map, IPI_STOP));
}

#if X86
int
resume_cpus(cpuset_t map)
{

	return (generic_restart_cpus(map, IPI_SUSPEND));
}
#endif
#undef X86

/*
 * All-CPU rendezvous.  CPUs are signalled, all execute the setup function 
 * (if specified), rendezvous, execute the action function (if specified),
 * rendezvous again, execute the teardown function (if specified), and then
 * resume.
 *
 * Note that the supplied external functions _must_ be reentrant and aware
 * that they are running in parallel and in an unknown lock context.
 */
void
smp_rendezvous_action(void)
{
	struct thread *td;
	void *local_func_arg;
	void (*local_setup_func)(void*);
	void (*local_action_func)(void*);
	void (*local_teardown_func)(void*);
#ifdef INVARIANTS
	int owepreempt;
#endif

	/* Ensure we have up-to-date values. */
	atomic_add_acq_int(&smp_rv_waiters[0], 1);
	while (smp_rv_waiters[0] < smp_rv_ncpus)
		cpu_spinwait();

	/* Fetch rendezvous parameters after acquire barrier. */
	local_func_arg = smp_rv_func_arg;
	local_setup_func = smp_rv_setup_func;
	local_action_func = smp_rv_action_func;
	local_teardown_func = smp_rv_teardown_func;

	/*
	 * Use a nested critical section to prevent any preemptions
	 * from occurring during a rendezvous action routine.
	 * Specifically, if a rendezvous handler is invoked via an IPI
	 * and the interrupted thread was in the critical_exit()
	 * function after setting td_critnest to 0 but before
	 * performing a deferred preemption, this routine can be
	 * invoked with td_critnest set to 0 and td_owepreempt true.
	 * In that case, a critical_exit() during the rendezvous
	 * action would trigger a preemption which is not permitted in
	 * a rendezvous action.  To fix this, wrap all of the
	 * rendezvous action handlers in a critical section.  We
	 * cannot use a regular critical section however as having
	 * critical_exit() preempt from this routine would also be
	 * problematic (the preemption must not occur before the IPI
	 * has been acknowledged via an EOI).  Instead, we
	 * intentionally ignore td_owepreempt when leaving the
	 * critical section.  This should be harmless because we do
	 * not permit rendezvous action routines to schedule threads,
	 * and thus td_owepreempt should never transition from 0 to 1
	 * during this routine.
	 */
	td = curthread;
	td->td_critnest++;
#ifdef INVARIANTS
	owepreempt = td->td_owepreempt;
#endif

	/*
	 * If requested, run a setup function before the main action
	 * function.  Ensure all CPUs have completed the setup
	 * function before moving on to the action function.
	 */
	if (local_setup_func != smp_no_rendezvous_barrier) {
		if (smp_rv_setup_func != NULL)
			smp_rv_setup_func(smp_rv_func_arg);
		atomic_add_int(&smp_rv_waiters[1], 1);
		while (smp_rv_waiters[1] < smp_rv_ncpus)
                	cpu_spinwait();
	}

	if (local_action_func != NULL)
		local_action_func(local_func_arg);

	if (local_teardown_func != smp_no_rendezvous_barrier) {
		/*
		 * Signal that the main action has been completed.  If a
		 * full exit rendezvous is requested, then all CPUs will
		 * wait here until all CPUs have finished the main action.
		 */
		atomic_add_int(&smp_rv_waiters[2], 1);
		while (smp_rv_waiters[2] < smp_rv_ncpus)
			cpu_spinwait();

		if (local_teardown_func != NULL)
			local_teardown_func(local_func_arg);
	}

	/*
	 * Signal that the rendezvous is fully completed by this CPU.
	 * This means that no member of smp_rv_* pseudo-structure will be
	 * accessed by this target CPU after this point; in particular,
	 * memory pointed by smp_rv_func_arg.
	 *
	 * The release semantic ensures that all accesses performed by
	 * the current CPU are visible when smp_rendezvous_cpus()
	 * returns, by synchronizing with the
	 * atomic_load_acq_int(&smp_rv_waiters[3]).
	 */
	atomic_add_rel_int(&smp_rv_waiters[3], 1);

	td->td_critnest--;
	KASSERT(owepreempt == td->td_owepreempt,
	    ("rendezvous action changed td_owepreempt"));
}

void
smp_rendezvous_cpus(cpuset_t map,
	void (* setup_func)(void *), 
	void (* action_func)(void *),
	void (* teardown_func)(void *),
	void *arg)
{
	int curcpumap, i, ncpus = 0;

	/* See comments in the !SMP case. */
	if (!smp_started) {
		spinlock_enter();
		if (setup_func != NULL)
			setup_func(arg);
		if (action_func != NULL)
			action_func(arg);
		if (teardown_func != NULL)
			teardown_func(arg);
		spinlock_exit();
		return;
	}

	/*
	 * Make sure we come here with interrupts enabled.  Otherwise we
	 * livelock if smp_ipi_mtx is owned by a thread which sent us an IPI.
	 */
	MPASS(curthread->td_md.md_spinlock_count == 0);

	CPU_FOREACH(i) {
		if (CPU_ISSET(i, &map))
			ncpus++;
	}
	if (ncpus == 0)
		panic("ncpus is 0 with non-zero map");

	mtx_lock_spin(&smp_ipi_mtx);

	/* Pass rendezvous parameters via global variables. */
	smp_rv_ncpus = ncpus;
	smp_rv_setup_func = setup_func;
	smp_rv_action_func = action_func;
	smp_rv_teardown_func = teardown_func;
	smp_rv_func_arg = arg;
	smp_rv_waiters[1] = 0;
	smp_rv_waiters[2] = 0;
	smp_rv_waiters[3] = 0;
	atomic_store_rel_int(&smp_rv_waiters[0], 0);

	/*
	 * Signal other processors, which will enter the IPI with
	 * interrupts off.
	 */
	curcpumap = CPU_ISSET(curcpu, &map);
	CPU_CLR(curcpu, &map);
	ipi_selected(map, IPI_RENDEZVOUS);

	/* Check if the current CPU is in the map */
	if (curcpumap != 0)
		smp_rendezvous_action();

	/*
	 * Ensure that the master CPU waits for all the other
	 * CPUs to finish the rendezvous, so that smp_rv_*
	 * pseudo-structure and the arg are guaranteed to not
	 * be in use.
	 *
	 * Load acquire synchronizes with the release add in
	 * smp_rendezvous_action(), which ensures that our caller sees
	 * all memory actions done by the called functions on other
	 * CPUs.
	 */
	while (atomic_load_acq_int(&smp_rv_waiters[3]) < ncpus)
		cpu_spinwait();

	mtx_unlock_spin(&smp_ipi_mtx);
}

void
smp_rendezvous(void (* setup_func)(void *), 
	       void (* action_func)(void *),
	       void (* teardown_func)(void *),
	       void *arg)
{
	smp_rendezvous_cpus(all_cpus, setup_func, action_func, teardown_func, arg);
}

static struct cpu_group group[MAXCPU * MAX_CACHE_LEVELS + 1];

struct cpu_group *
smp_topo(void)
{
	char cpusetbuf[CPUSETBUFSIZ], cpusetbuf2[CPUSETBUFSIZ];
	struct cpu_group *top;

	/*
	 * Check for a fake topology request for debugging purposes.
	 */
	switch (smp_topology) {
	case 1:
		/* Dual core with no sharing.  */
		top = smp_topo_1level(CG_SHARE_NONE, 2, 0);
		break;
	case 2:
		/* No topology, all cpus are equal. */
		top = smp_topo_none();
		break;
	case 3:
		/* Dual core with shared L2.  */
		top = smp_topo_1level(CG_SHARE_L2, 2, 0);
		break;
	case 4:
		/* quad core, shared l3 among each package, private l2.  */
		top = smp_topo_1level(CG_SHARE_L3, 4, 0);
		break;
	case 5:
		/* quad core,  2 dualcore parts on each package share l2.  */
		top = smp_topo_2level(CG_SHARE_NONE, 2, CG_SHARE_L2, 2, 0);
		break;
	case 6:
		/* Single-core 2xHTT */
		top = smp_topo_1level(CG_SHARE_L1, 2, CG_FLAG_HTT);
		break;
	case 7:
		/* quad core with a shared l3, 8 threads sharing L2.  */
		top = smp_topo_2level(CG_SHARE_L3, 4, CG_SHARE_L2, 8,
		    CG_FLAG_SMT);
		break;
	default:
		/* Default, ask the system what it wants. */
		top = cpu_topo();
		break;
	}
	/*
	 * Verify the returned topology.
	 */
	if (top->cg_count != mp_ncpus)
		panic("Built bad topology at %p.  CPU count %d != %d",
		    top, top->cg_count, mp_ncpus);
	if (CPU_CMP(&top->cg_mask, &all_cpus))
		panic("Built bad topology at %p.  CPU mask (%s) != (%s)",
		    top, cpusetobj_strprint(cpusetbuf, &top->cg_mask),
		    cpusetobj_strprint(cpusetbuf2, &all_cpus));

	/*
	 * Collapse nonsense levels that may be created out of convenience by
	 * the MD layers.  They cause extra work in the search functions.
	 */
	while (top->cg_children == 1) {
		top = &top->cg_child[0];
		top->cg_parent = NULL;
	}
	return (top);
}

struct cpu_group *
smp_topo_alloc(u_int count)
{
	static u_int index;
	u_int curr;

	curr = index;
	index += count;
	return (&group[curr]);
}

struct cpu_group *
smp_topo_none(void)
{
	struct cpu_group *top;

	top = &group[0];
	top->cg_parent = NULL;
	top->cg_child = NULL;
	top->cg_mask = all_cpus;
	top->cg_count = mp_ncpus;
	top->cg_children = 0;
	top->cg_level = CG_SHARE_NONE;
	top->cg_flags = 0;

	return (top);
}

static int
smp_topo_addleaf(struct cpu_group *parent, struct cpu_group *child, int share,
    int count, int flags, int start)
{
	char cpusetbuf[CPUSETBUFSIZ], cpusetbuf2[CPUSETBUFSIZ];
	cpuset_t mask;
	int i;

	CPU_ZERO(&mask);
	for (i = 0; i < count; i++, start++)
		CPU_SET(start, &mask);
	child->cg_parent = parent;
	child->cg_child = NULL;
	child->cg_children = 0;
	child->cg_level = share;
	child->cg_count = count;
	child->cg_flags = flags;
	child->cg_mask = mask;
	parent->cg_children++;
	for (; parent != NULL; parent = parent->cg_parent) {
		if (CPU_OVERLAP(&parent->cg_mask, &child->cg_mask))
			panic("Duplicate children in %p.  mask (%s) child (%s)",
			    parent,
			    cpusetobj_strprint(cpusetbuf, &parent->cg_mask),
			    cpusetobj_strprint(cpusetbuf2, &child->cg_mask));
		CPU_OR(&parent->cg_mask, &child->cg_mask);
		parent->cg_count += child->cg_count;
	}

	return (start);
}

struct cpu_group *
smp_topo_1level(int share, int count, int flags)
{
	struct cpu_group *child;
	struct cpu_group *top;
	int packages;
	int cpu;
	int i;

	cpu = 0;
	top = &group[0];
	packages = mp_ncpus / count;
	top->cg_child = child = &group[1];
	top->cg_level = CG_SHARE_NONE;
	for (i = 0; i < packages; i++, child++)
		cpu = smp_topo_addleaf(top, child, share, count, flags, cpu);
	return (top);
}

struct cpu_group *
smp_topo_2level(int l2share, int l2count, int l1share, int l1count,
    int l1flags)
{
	struct cpu_group *top;
	struct cpu_group *l1g;
	struct cpu_group *l2g;
	int cpu;
	int i;
	int j;

	cpu = 0;
	top = &group[0];
	l2g = &group[1];
	top->cg_child = l2g;
	top->cg_level = CG_SHARE_NONE;
	top->cg_children = mp_ncpus / (l2count * l1count);
	l1g = l2g + top->cg_children;
	for (i = 0; i < top->cg_children; i++, l2g++) {
		l2g->cg_parent = top;
		l2g->cg_child = l1g;
		l2g->cg_level = l2share;
		for (j = 0; j < l2count; j++, l1g++)
			cpu = smp_topo_addleaf(l2g, l1g, l1share, l1count,
			    l1flags, cpu);
	}
	return (top);
}

struct cpu_group *
smp_topo_find(struct cpu_group *top, int cpu)
{
	struct cpu_group *cg;
	cpuset_t mask;
	int children;
	int i;

	CPU_SETOF(cpu, &mask);
	cg = top;
	for (;;) {
		if (!CPU_OVERLAP(&cg->cg_mask, &mask))
			return (NULL);
		if (cg->cg_children == 0)
			return (cg);
		children = cg->cg_children;
		for (i = 0, cg = cg->cg_child; i < children; cg++, i++)
			if (CPU_OVERLAP(&cg->cg_mask, &mask))
				break;
	}
	return (NULL);
}
#else /* !SMP */

void
smp_rendezvous_cpus(cpuset_t map,
	void (*setup_func)(void *), 
	void (*action_func)(void *),
	void (*teardown_func)(void *),
	void *arg)
{
	/*
	 * In the !SMP case we just need to ensure the same initial conditions
	 * as the SMP case.
	 */
	spinlock_enter();
	if (setup_func != NULL)
		setup_func(arg);
	if (action_func != NULL)
		action_func(arg);
	if (teardown_func != NULL)
		teardown_func(arg);
	spinlock_exit();
}

void
smp_rendezvous(void (*setup_func)(void *), 
	       void (*action_func)(void *),
	       void (*teardown_func)(void *),
	       void *arg)
{

	smp_rendezvous_cpus(all_cpus, setup_func, action_func, teardown_func,
	    arg);
}

/*
 * Provide dummy SMP support for UP kernels.  Modules that need to use SMP
 * APIs will still work using this dummy support.
 */
static void
mp_setvariables_for_up(void *dummy)
{
	mp_ncpus = 1;
	mp_ncores = 1;
	mp_maxid = PCPU_GET(cpuid);
	CPU_SETOF(mp_maxid, &all_cpus);
	KASSERT(PCPU_GET(cpuid) == 0, ("UP must have a CPU ID of zero"));
}
SYSINIT(cpu_mp_setvariables, SI_SUB_TUNABLES, SI_ORDER_FIRST,
    mp_setvariables_for_up, NULL);
#endif /* SMP */

void
smp_no_rendezvous_barrier(void *dummy)
{
#ifdef SMP
	KASSERT((!smp_started),("smp_no_rendezvous called and smp is started"));
#endif
}

void
smp_rendezvous_cpus_retry(cpuset_t map,
	void (* setup_func)(void *),
	void (* action_func)(void *),
	void (* teardown_func)(void *),
	void (* wait_func)(void *, int),
	struct smp_rendezvous_cpus_retry_arg *arg)
{
	int cpu;

	CPU_COPY(&map, &arg->cpus);

	/*
	 * Only one CPU to execute on.
	 */
	if (!smp_started) {
		spinlock_enter();
		if (setup_func != NULL)
			setup_func(arg);
		if (action_func != NULL)
			action_func(arg);
		if (teardown_func != NULL)
			teardown_func(arg);
		spinlock_exit();
		return;
	}

	/*
	 * Execute an action on all specified CPUs while retrying until they
	 * all acknowledge completion.
	 */
	for (;;) {
		smp_rendezvous_cpus(
		    arg->cpus,
		    setup_func,
		    action_func,
		    teardown_func,
		    arg);

		if (CPU_EMPTY(&arg->cpus))
			break;

		CPU_FOREACH(cpu) {
			if (!CPU_ISSET(cpu, &arg->cpus))
				continue;
			wait_func(arg, cpu);
		}
	}
}

void
smp_rendezvous_cpus_done(struct smp_rendezvous_cpus_retry_arg *arg)
{

	CPU_CLR_ATOMIC(curcpu, &arg->cpus);
}

/*
 * If (prio & PDROP) == 0:
 * Wait for specified idle threads to switch once.  This ensures that even
 * preempted threads have cycled through the switch function once,
 * exiting their codepaths.  This allows us to change global pointers
 * with no other synchronization.
 * If (prio & PDROP) != 0:
 * Force the specified CPUs to switch context at least once.
 */
int
quiesce_cpus(cpuset_t map, const char *wmesg, int prio)
{
	struct pcpu *pcpu;
	u_int *gen;
	int error;
	int cpu;

	error = 0;
	if ((prio & PDROP) == 0) {
		gen = malloc(sizeof(u_int) * MAXCPU, M_TEMP, M_WAITOK);
		for (cpu = 0; cpu <= mp_maxid; cpu++) {
			if (!CPU_ISSET(cpu, &map) || CPU_ABSENT(cpu))
				continue;
			pcpu = pcpu_find(cpu);
			gen[cpu] = pcpu->pc_idlethread->td_generation;
		}
	}
	for (cpu = 0; cpu <= mp_maxid; cpu++) {
		if (!CPU_ISSET(cpu, &map) || CPU_ABSENT(cpu))
			continue;
		pcpu = pcpu_find(cpu);
		thread_lock(curthread);
		sched_bind(curthread, cpu);
		thread_unlock(curthread);
		if ((prio & PDROP) != 0)
			continue;
		while (gen[cpu] == pcpu->pc_idlethread->td_generation) {
			error = tsleep(quiesce_cpus, prio & ~PDROP, wmesg, 1);
			if (error != EWOULDBLOCK)
				goto out;
			error = 0;
		}
	}
out:
	thread_lock(curthread);
	sched_unbind(curthread);
	thread_unlock(curthread);
	if ((prio & PDROP) == 0)
		free(gen, M_TEMP);

	return (error);
}

int
quiesce_all_cpus(const char *wmesg, int prio)
{

	return quiesce_cpus(all_cpus, wmesg, prio);
}

/*
 * Observe all CPUs not executing in critical section.
 * We are not in one so the check for us is safe. If the found
 * thread changes to something else we know the section was
 * exited as well.
 */
void
quiesce_all_critical(void)
{
	struct thread *td, *newtd;
	struct pcpu *pcpu;
	int cpu;

	MPASS(curthread->td_critnest == 0);

	CPU_FOREACH(cpu) {
		pcpu = cpuid_to_pcpu[cpu];
		td = pcpu->pc_curthread;
		for (;;) {
			if (td->td_critnest == 0)
				break;
			cpu_spinwait();
			newtd = (struct thread *)
			    atomic_load_acq_ptr((void *)pcpu->pc_curthread);
			if (td != newtd)
				break;
		}
	}
}

static void
cpus_fence_seq_cst_issue(void *arg __unused)
{

	atomic_thread_fence_seq_cst();
}

/*
 * Send an IPI forcing a sequentially consistent fence.
 *
 * Allows replacement of an explicitly fence with a compiler barrier.
 * Trades speed up during normal execution for a significant slowdown when
 * the barrier is needed.
 */
void
cpus_fence_seq_cst(void)
{

#ifdef SMP
	smp_rendezvous(
	    smp_no_rendezvous_barrier,
	    cpus_fence_seq_cst_issue,
	    smp_no_rendezvous_barrier,
	    NULL
	);
#else
	cpus_fence_seq_cst_issue(NULL);
#endif
}

/* Extra care is taken with this sysctl because the data type is volatile */
static int
sysctl_kern_smp_active(SYSCTL_HANDLER_ARGS)
{
	int error, active;

	active = smp_started;
	error = SYSCTL_OUT(req, &active, sizeof(active));
	return (error);
}

#ifdef SMP
void
topo_init_node(struct topo_node *node)
{

	bzero(node, sizeof(*node));
	TAILQ_INIT(&node->children);
}

void
topo_init_root(struct topo_node *root)
{

	topo_init_node(root);
	root->type = TOPO_TYPE_SYSTEM;
}

/*
 * Add a child node with the given ID under the given parent.
 * Do nothing if there is already a child with that ID.
 */
struct topo_node *
topo_add_node_by_hwid(struct topo_node *parent, int hwid,
    topo_node_type type, uintptr_t subtype)
{
	struct topo_node *node;

	TAILQ_FOREACH_REVERSE(node, &parent->children,
	    topo_children, siblings) {
		if (node->hwid == hwid
		    && node->type == type && node->subtype == subtype) {
			return (node);
		}
	}

	node = malloc(sizeof(*node), M_TOPO, M_WAITOK);
	topo_init_node(node);
	node->parent = parent;
	node->hwid = hwid;
	node->type = type;
	node->subtype = subtype;
	TAILQ_INSERT_TAIL(&parent->children, node, siblings);
	parent->nchildren++;

	return (node);
}

/*
 * Find a child node with the given ID under the given parent.
 */
struct topo_node *
topo_find_node_by_hwid(struct topo_node *parent, int hwid,
    topo_node_type type, uintptr_t subtype)
{

	struct topo_node *node;

	TAILQ_FOREACH(node, &parent->children, siblings) {
		if (node->hwid == hwid
		    && node->type == type && node->subtype == subtype) {
			return (node);
		}
	}

	return (NULL);
}

/*
 * Given a node change the order of its parent's child nodes such
 * that the node becomes the firt child while preserving the cyclic
 * order of the children.  In other words, the given node is promoted
 * by rotation.
 */
void
topo_promote_child(struct topo_node *child)
{
	struct topo_node *next;
	struct topo_node *node;
	struct topo_node *parent;

	parent = child->parent;
	next = TAILQ_NEXT(child, siblings);
	TAILQ_REMOVE(&parent->children, child, siblings);
	TAILQ_INSERT_HEAD(&parent->children, child, siblings);

	while (next != NULL) {
		node = next;
		next = TAILQ_NEXT(node, siblings);
		TAILQ_REMOVE(&parent->children, node, siblings);
		TAILQ_INSERT_AFTER(&parent->children, child, node, siblings);
		child = node;
	}
}

/*
 * Iterate to the next node in the depth-first search (traversal) of
 * the topology tree.
 */
struct topo_node *
topo_next_node(struct topo_node *top, struct topo_node *node)
{
	struct topo_node *next;

	if ((next = TAILQ_FIRST(&node->children)) != NULL)
		return (next);

	if ((next = TAILQ_NEXT(node, siblings)) != NULL)
		return (next);

	while (node != top && (node = node->parent) != top)
		if ((next = TAILQ_NEXT(node, siblings)) != NULL)
			return (next);

	return (NULL);
}

/*
 * Iterate to the next node in the depth-first search of the topology tree,
 * but without descending below the current node.
 */
struct topo_node *
topo_next_nonchild_node(struct topo_node *top, struct topo_node *node)
{
	struct topo_node *next;

	if ((next = TAILQ_NEXT(node, siblings)) != NULL)
		return (next);

	while (node != top && (node = node->parent) != top)
		if ((next = TAILQ_NEXT(node, siblings)) != NULL)
			return (next);

	return (NULL);
}

/*
 * Assign the given ID to the given topology node that represents a logical
 * processor.
 */
void
topo_set_pu_id(struct topo_node *node, cpuid_t id)
{

	KASSERT(node->type == TOPO_TYPE_PU,
	    ("topo_set_pu_id: wrong node type: %u", node->type));
	KASSERT(CPU_EMPTY(&node->cpuset) && node->cpu_count == 0,
	    ("topo_set_pu_id: cpuset already not empty"));
	node->id = id;
	CPU_SET(id, &node->cpuset);
	node->cpu_count = 1;
	node->subtype = 1;

	while ((node = node->parent) != NULL) {
		KASSERT(!CPU_ISSET(id, &node->cpuset),
		    ("logical ID %u is already set in node %p", id, node));
		CPU_SET(id, &node->cpuset);
		node->cpu_count++;
	}
}

static struct topology_spec {
	topo_node_type	type;
	bool		match_subtype;
	uintptr_t	subtype;
} topology_level_table[TOPO_LEVEL_COUNT] = {
	[TOPO_LEVEL_PKG] = { .type = TOPO_TYPE_PKG, },
	[TOPO_LEVEL_GROUP] = { .type = TOPO_TYPE_GROUP, },
	[TOPO_LEVEL_CACHEGROUP] = {
		.type = TOPO_TYPE_CACHE,
		.match_subtype = true,
		.subtype = CG_SHARE_L3,
	},
	[TOPO_LEVEL_CORE] = { .type = TOPO_TYPE_CORE, },
	[TOPO_LEVEL_THREAD] = { .type = TOPO_TYPE_PU, },
};

static bool
topo_analyze_table(struct topo_node *root, int all, enum topo_level level,
    struct topo_analysis *results)
{
	struct topology_spec *spec;
	struct topo_node *node;
	int count;

	if (level >= TOPO_LEVEL_COUNT)
		return (true);

	spec = &topology_level_table[level];
	count = 0;
	node = topo_next_node(root, root);

	while (node != NULL) {
		if (node->type != spec->type ||
		    (spec->match_subtype && node->subtype != spec->subtype)) {
			node = topo_next_node(root, node);
			continue;
		}
		if (!all && CPU_EMPTY(&node->cpuset)) {
			node = topo_next_nonchild_node(root, node);
			continue;
		}

		count++;

		if (!topo_analyze_table(node, all, level + 1, results))
			return (false);

		node = topo_next_nonchild_node(root, node);
	}

	/* No explicit subgroups is essentially one subgroup. */
	if (count == 0) {
		count = 1;

		if (!topo_analyze_table(root, all, level + 1, results))
			return (false);
	}

	if (results->entities[level] == -1)
		results->entities[level] = count;
	else if (results->entities[level] != count)
		return (false);

	return (true);
}

/*
 * Check if the topology is uniform, that is, each package has the same number
 * of cores in it and each core has the same number of threads (logical
 * processors) in it.  If so, calculate the number of packages, the number of
 * groups per package, the number of cachegroups per group, and the number of
 * logical processors per cachegroup.  'all' parameter tells whether to include
 * administratively disabled logical processors into the analysis.
 */
int
topo_analyze(struct topo_node *topo_root, int all,
    struct topo_analysis *results)
{

	results->entities[TOPO_LEVEL_PKG] = -1;
	results->entities[TOPO_LEVEL_CORE] = -1;
	results->entities[TOPO_LEVEL_THREAD] = -1;
	results->entities[TOPO_LEVEL_GROUP] = -1;
	results->entities[TOPO_LEVEL_CACHEGROUP] = -1;

	if (!topo_analyze_table(topo_root, all, TOPO_LEVEL_PKG, results))
		return (0);

	KASSERT(results->entities[TOPO_LEVEL_PKG] > 0,
		("bug in topology or analysis"));

	return (1);
}

#endif /* SMP */