aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/rcupreempt.c
blob: ca4bbbe04aa4db9d150a503c9025dbe83e0d62a3 (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
/*
 * Read-Copy Update mechanism for mutual exclusion, realtime implementation
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
 *
 * Copyright IBM Corporation, 2006
 *
 * Authors: Paul E. McKenney <paulmck@us.ibm.com>
 *		With thanks to Esben Nielsen, Bill Huey, and Ingo Molnar
 *		for pushing me away from locks and towards counters, and
 *		to Suparna Bhattacharya for pushing me completely away
 *		from atomic instructions on the read side.
 *
 *  - Added handling of Dynamic Ticks
 *      Copyright 2007 - Paul E. Mckenney <paulmck@us.ibm.com>
 *                     - Steven Rostedt <srostedt@redhat.com>
 *
 * Papers:  http://www.rdrop.com/users/paulmck/RCU
 *
 * Design Document: http://lwn.net/Articles/253651/
 *
 * For detailed explanation of Read-Copy Update mechanism see -
 * 		Documentation/RCU/ *.txt
 *
 */
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/spinlock.h>
#include <linux/smp.h>
#include <linux/rcupdate.h>
#include <linux/interrupt.h>
#include <linux/sched.h>
#include <asm/atomic.h>
#include <linux/bitops.h>
#include <linux/module.h>
#include <linux/kthread.h>
#include <linux/completion.h>
#include <linux/moduleparam.h>
#include <linux/percpu.h>
#include <linux/notifier.h>
#include <linux/cpu.h>
#include <linux/random.h>
#include <linux/delay.h>
#include <linux/byteorder/swabb.h>
#include <linux/cpumask.h>
#include <linux/rcupreempt_trace.h>

/*
 * PREEMPT_RCU data structures.
 */

/*
 * GP_STAGES specifies the number of times the state machine has
 * to go through the all the rcu_try_flip_states (see below)
 * in a single Grace Period.
 *
 * GP in GP_STAGES stands for Grace Period ;)
 */
#define GP_STAGES    2
struct rcu_data {
	spinlock_t	lock;		/* Protect rcu_data fields. */
	long		completed;	/* Number of last completed batch. */
	int		waitlistcount;
	struct rcu_head *nextlist;
	struct rcu_head **nexttail;
	struct rcu_head *waitlist[GP_STAGES];
	struct rcu_head **waittail[GP_STAGES];
	struct rcu_head *donelist;	/* from waitlist & waitschedlist */
	struct rcu_head **donetail;
	long rcu_flipctr[2];
	struct rcu_head *nextschedlist;
	struct rcu_head **nextschedtail;
	struct rcu_head *waitschedlist;
	struct rcu_head **waitschedtail;
	int rcu_sched_sleeping;
#ifdef CONFIG_RCU_TRACE
	struct rcupreempt_trace trace;
#endif /* #ifdef CONFIG_RCU_TRACE */
};

/*
 * States for rcu_try_flip() and friends.
 */

enum rcu_try_flip_states {

	/*
	 * Stay here if nothing is happening. Flip the counter if somthing
	 * starts happening. Denoted by "I"
	 */
	rcu_try_flip_idle_state,

	/*
	 * Wait here for all CPUs to notice that the counter has flipped. This
	 * prevents the old set of counters from ever being incremented once
	 * we leave this state, which in turn is necessary because we cannot
	 * test any individual counter for zero -- we can only check the sum.
	 * Denoted by "A".
	 */
	rcu_try_flip_waitack_state,

	/*
	 * Wait here for the sum of the old per-CPU counters to reach zero.
	 * Denoted by "Z".
	 */
	rcu_try_flip_waitzero_state,

	/*
	 * Wait here for each of the other CPUs to execute a memory barrier.
	 * This is necessary to ensure that these other CPUs really have
	 * completed executing their RCU read-side critical sections, despite
	 * their CPUs wildly reordering memory. Denoted by "M".
	 */
	rcu_try_flip_waitmb_state,
};

/*
 * States for rcu_ctrlblk.rcu_sched_sleep.
 */

enum rcu_sched_sleep_states {
	rcu_sched_not_sleeping,	/* Not sleeping, callbacks need GP.  */
	rcu_sched_sleep_prep,	/* Thinking of sleeping, rechecking. */
	rcu_sched_sleeping,	/* Sleeping, awaken if GP needed. */
};

struct rcu_ctrlblk {
	spinlock_t	fliplock;	/* Protect state-machine transitions. */
	long		completed;	/* Number of last completed batch. */
	enum rcu_try_flip_states rcu_try_flip_state; /* The current state of
							the rcu state machine */
	spinlock_t	schedlock;	/* Protect rcu_sched sleep state. */
	enum rcu_sched_sleep_states sched_sleep; /* rcu_sched state. */
	wait_queue_head_t sched_wq;	/* Place for rcu_sched to sleep. */
};

static DEFINE_PER_CPU(struct rcu_data, rcu_data);
static struct rcu_ctrlblk rcu_ctrlblk = {
	.fliplock = __SPIN_LOCK_UNLOCKED(rcu_ctrlblk.fliplock),
	.completed = 0,
	.rcu_try_flip_state = rcu_try_flip_idle_state,
	.schedlock = __SPIN_LOCK_UNLOCKED(rcu_ctrlblk.schedlock),
	.sched_sleep = rcu_sched_not_sleeping,
	.sched_wq = __WAIT_QUEUE_HEAD_INITIALIZER(rcu_ctrlblk.sched_wq),
};

static struct task_struct *rcu_sched_grace_period_task;

#ifdef CONFIG_RCU_TRACE
static char *rcu_try_flip_state_names[] =
	{ "idle", "waitack", "waitzero", "waitmb" };
#endif /* #ifdef CONFIG_RCU_TRACE */

static cpumask_t rcu_cpu_online_map __read_mostly = CPU_MASK_NONE;

/*
 * Enum and per-CPU flag to determine when each CPU has seen
 * the most recent counter flip.
 */

enum rcu_flip_flag_values {
	rcu_flip_seen,		/* Steady/initial state, last flip seen. */
				/* Only GP detector can update. */
	rcu_flipped		/* Flip just completed, need confirmation. */
				/* Only corresponding CPU can update. */
};
static DEFINE_PER_CPU_SHARED_ALIGNED(enum rcu_flip_flag_values, rcu_flip_flag)
								= rcu_flip_seen;

/*
 * Enum and per-CPU flag to determine when each CPU has executed the
 * needed memory barrier to fence in memory references from its last RCU
 * read-side critical section in the just-completed grace period.
 */

enum rcu_mb_flag_values {
	rcu_mb_done,		/* Steady/initial state, no mb()s required. */
				/* Only GP detector can update. */
	rcu_mb_needed		/* Flip just completed, need an mb(). */
				/* Only corresponding CPU can update. */
};
static DEFINE_PER_CPU_SHARED_ALIGNED(enum rcu_mb_flag_values, rcu_mb_flag)
								= rcu_mb_done;

/*
 * RCU_DATA_ME: find the current CPU's rcu_data structure.
 * RCU_DATA_CPU: find the specified CPU's rcu_data structure.
 */
#define RCU_DATA_ME()		(&__get_cpu_var(rcu_data))
#define RCU_DATA_CPU(cpu)	(&per_cpu(rcu_data, cpu))

/*
 * Helper macro for tracing when the appropriate rcu_data is not
 * cached in a local variable, but where the CPU number is so cached.
 */
#define RCU_TRACE_CPU(f, cpu) RCU_TRACE(f, &(RCU_DATA_CPU(cpu)->trace));

/*
 * Helper macro for tracing when the appropriate rcu_data is not
 * cached in a local variable.
 */
#define RCU_TRACE_ME(f) RCU_TRACE(f, &(RCU_DATA_ME()->trace));

/*
 * Helper macro for tracing when the appropriate rcu_data is pointed
 * to by a local variable.
 */
#define RCU_TRACE_RDP(f, rdp) RCU_TRACE(f, &((rdp)->trace));

#define RCU_SCHED_BATCH_TIME (HZ / 50)

/*
 * Return the number of RCU batches processed thus far.  Useful
 * for debug and statistics.
 */
long rcu_batches_completed(void)
{
	return rcu_ctrlblk.completed;
}
EXPORT_SYMBOL_GPL(rcu_batches_completed);

void __rcu_read_lock(void)
{
	int idx;
	struct task_struct *t = current;
	int nesting;

	nesting = ACCESS_ONCE(t->rcu_read_lock_nesting);
	if (nesting != 0) {

		/* An earlier rcu_read_lock() covers us, just count it. */

		t->rcu_read_lock_nesting = nesting + 1;

	} else {
		unsigned long flags;

		/*
		 * We disable interrupts for the following reasons:
		 * - If we get scheduling clock interrupt here, and we
		 *   end up acking the counter flip, it's like a promise
		 *   that we will never increment the old counter again.
		 *   Thus we will break that promise if that
		 *   scheduling clock interrupt happens between the time
		 *   we pick the .completed field and the time that we
		 *   increment our counter.
		 *
		 * - We don't want to be preempted out here.
		 *
		 * NMIs can still occur, of course, and might themselves
		 * contain rcu_read_lock().
		 */

		local_irq_save(flags);

		/*
		 * Outermost nesting of rcu_read_lock(), so increment
		 * the current counter for the current CPU.  Use volatile
		 * casts to prevent the compiler from reordering.
		 */

		idx = ACCESS_ONCE(rcu_ctrlblk.completed) & 0x1;
		ACCESS_ONCE(RCU_DATA_ME()->rcu_flipctr[idx])++;

		/*
		 * Now that the per-CPU counter has been incremented, we
		 * are protected from races with rcu_read_lock() invoked
		 * from NMI handlers on this CPU.  We can therefore safely
		 * increment the nesting counter, relieving further NMIs
		 * of the need to increment the per-CPU counter.
		 */

		ACCESS_ONCE(t->rcu_read_lock_nesting) = nesting + 1;

		/*
		 * Now that we have preventing any NMIs from storing
		 * to the ->rcu_flipctr_idx, we can safely use it to
		 * remember which counter to decrement in the matching
		 * rcu_read_unlock().
		 */

		ACCESS_ONCE(t->rcu_flipctr_idx) = idx;
		local_irq_restore(flags);
	}
}
EXPORT_SYMBOL_GPL(__rcu_read_lock);

void __rcu_read_unlock(void)
{
	int idx;
	struct task_struct *t = current;
	int nesting;

	nesting = ACCESS_ONCE(t->rcu_read_lock_nesting);
	if (nesting > 1) {

		/*
		 * We are still protected by the enclosing rcu_read_lock(),
		 * so simply decrement the counter.
		 */

		t->rcu_read_lock_nesting = nesting - 1;

	} else {
		unsigned long flags;

		/*
		 * Disable local interrupts to prevent the grace-period
		 * detection state machine from seeing us half-done.
		 * NMIs can still occur, of course, and might themselves
		 * contain rcu_read_lock() and rcu_read_unlock().
		 */

		local_irq_save(flags);

		/*
		 * Outermost nesting of rcu_read_unlock(), so we must
		 * decrement the current counter for the current CPU.
		 * This must be done carefully, because NMIs can
		 * occur at any point in this code, and any rcu_read_lock()
		 * and rcu_read_unlock() pairs in the NMI handlers
		 * must interact non-destructively with this code.
		 * Lots of volatile casts, and -very- careful ordering.
		 *
		 * Changes to this code, including this one, must be
		 * inspected, validated, and tested extremely carefully!!!
		 */

		/*
		 * First, pick up the index.
		 */

		idx = ACCESS_ONCE(t->rcu_flipctr_idx);

		/*
		 * Now that we have fetched the counter index, it is
		 * safe to decrement the per-task RCU nesting counter.
		 * After this, any interrupts or NMIs will increment and
		 * decrement the per-CPU counters.
		 */
		ACCESS_ONCE(t->rcu_read_lock_nesting) = nesting - 1;

		/*
		 * It is now safe to decrement this task's nesting count.
		 * NMIs that occur after this statement will route their
		 * rcu_read_lock() calls through this "else" clause, and
		 * will thus start incrementing the per-CPU counter on
		 * their own.  They will also clobber ->rcu_flipctr_idx,
		 * but that is OK, since we have already fetched it.
		 */

		ACCESS_ONCE(RCU_DATA_ME()->rcu_flipctr[idx])--;
		local_irq_restore(flags);
	}
}
EXPORT_SYMBOL_GPL(__rcu_read_unlock);

/*
 * If a global counter flip has occurred since the last time that we
 * advanced callbacks, advance them.  Hardware interrupts must be
 * disabled when calling this function.
 */
static void __rcu_advance_callbacks(struct rcu_data *rdp)
{
	int cpu;
	int i;
	int wlc = 0;

	if (rdp->completed != rcu_ctrlblk.completed) {
		if (rdp->waitlist[GP_STAGES - 1] != NULL) {
			*rdp->donetail = rdp->waitlist[GP_STAGES - 1];
			rdp->donetail = rdp->waittail[GP_STAGES - 1];
			RCU_TRACE_RDP(rcupreempt_trace_move2done, rdp);
		}
		for (i = GP_STAGES - 2; i >= 0; i--) {
			if (rdp->waitlist[i] != NULL) {
				rdp->waitlist[i + 1] = rdp->waitlist[i];
				rdp->waittail[i + 1] = rdp->waittail[i];
				wlc++;
			} else {
				rdp->waitlist[i + 1] = NULL;
				rdp->waittail[i + 1] =
					&rdp->waitlist[i + 1];
			}
		}
		if (rdp->nextlist != NULL) {
			rdp->waitlist[0] = rdp->nextlist;
			rdp->waittail[0] = rdp->nexttail;
			wlc++;
			rdp->nextlist = NULL;
			rdp->nexttail = &rdp->nextlist;
			RCU_TRACE_RDP(rcupreempt_trace_move2wait, rdp);
		} else {
			rdp->waitlist[0] = NULL;
			rdp->waittail[0] = &rdp->waitlist[0];
		}
		rdp->waitlistcount = wlc;
		rdp->completed = rcu_ctrlblk.completed;
	}

	/*
	 * Check to see if this CPU needs to report that it has seen
	 * the most recent counter flip, thereby declaring that all
	 * subsequent rcu_read_lock() invocations will respect this flip.
	 */

	cpu = raw_smp_processor_id();
	if (per_cpu(rcu_flip_flag, cpu) == rcu_flipped) {
		smp_mb();  /* Subsequent counter accesses must see new value */
		per_cpu(rcu_flip_flag, cpu) = rcu_flip_seen;
		smp_mb();  /* Subsequent RCU read-side critical sections */
			   /*  seen -after- acknowledgement. */
	}
}

DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_dyntick_sched, rcu_dyntick_sched) = {
	.dynticks = 1,
};

#ifdef CONFIG_NO_HZ
static DEFINE_PER_CPU(int, rcu_update_flag);

/**
 * rcu_irq_enter - Called from Hard irq handlers and NMI/SMI.
 *
 * If the CPU was idle with dynamic ticks active, this updates the
 * rcu_dyntick_sched.dynticks to let the RCU handling know that the
 * CPU is active.
 */
void rcu_irq_enter(void)
{
	int cpu = smp_processor_id();
	struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);

	if (per_cpu(rcu_update_flag, cpu))
		per_cpu(rcu_update_flag, cpu)++;

	/*
	 * Only update if we are coming from a stopped ticks mode
	 * (rcu_dyntick_sched.dynticks is even).
	 */
	if (!in_interrupt() &&
	    (rdssp->dynticks & 0x1) == 0) {
		/*
		 * The following might seem like we could have a race
		 * with NMI/SMIs. But this really isn't a problem.
		 * Here we do a read/modify/write, and the race happens
		 * when an NMI/SMI comes in after the read and before
		 * the write. But NMI/SMIs will increment this counter
		 * twice before returning, so the zero bit will not
		 * be corrupted by the NMI/SMI which is the most important
		 * part.
		 *
		 * The only thing is that we would bring back the counter
		 * to a postion that it was in during the NMI/SMI.
		 * But the zero bit would be set, so the rest of the
		 * counter would again be ignored.
		 *
		 * On return from the IRQ, the counter may have the zero
		 * bit be 0 and the counter the same as the return from
		 * the NMI/SMI. If the state machine was so unlucky to
		 * see that, it still doesn't matter, since all
		 * RCU read-side critical sections on this CPU would
		 * have already completed.
		 */
		rdssp->dynticks++;
		/*
		 * The following memory barrier ensures that any
		 * rcu_read_lock() primitives in the irq handler
		 * are seen by other CPUs to follow the above
		 * increment to rcu_dyntick_sched.dynticks. This is
		 * required in order for other CPUs to correctly
		 * determine when it is safe to advance the RCU
		 * grace-period state machine.
		 */
		smp_mb(); /* see above block comment. */
		/*
		 * Since we can't determine the dynamic tick mode from
		 * the rcu_dyntick_sched.dynticks after this routine,
		 * we use a second flag to acknowledge that we came
		 * from an idle state with ticks stopped.
		 */
		per_cpu(rcu_update_flag, cpu)++;
		/*
		 * If we take an NMI/SMI now, they will also increment
		 * the rcu_update_flag, and will not update the
		 * rcu_dyntick_sched.dynticks on exit. That is for
		 * this IRQ to do.
		 */
	}
}

/**
 * rcu_irq_exit - Called from exiting Hard irq context.
 *
 * If the CPU was idle with dynamic ticks active, update the
 * rcu_dyntick_sched.dynticks to put let the RCU handling be
 * aware that the CPU is going back to idle with no ticks.
 */
void rcu_irq_exit(void)
{
	int cpu = smp_processor_id();
	struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);

	/*
	 * rcu_update_flag is set if we interrupted the CPU
	 * when it was idle with ticks stopped.
	 * Once this occurs, we keep track of interrupt nesting
	 * because a NMI/SMI could also come in, and we still
	 * only want the IRQ that started the increment of the
	 * rcu_dyntick_sched.dynticks to be the one that modifies
	 * it on exit.
	 */
	if (per_cpu(rcu_update_flag, cpu)) {
		if (--per_cpu(rcu_update_flag, cpu))
			return;

		/* This must match the interrupt nesting */
		WARN_ON(in_interrupt());

		/*
		 * If an NMI/SMI happens now we are still
		 * protected by the rcu_dyntick_sched.dynticks being odd.
		 */

		/*
		 * The following memory barrier ensures that any
		 * rcu_read_unlock() primitives in the irq handler
		 * are seen by other CPUs to preceed the following
		 * increment to rcu_dyntick_sched.dynticks. This
		 * is required in order for other CPUs to determine
		 * when it is safe to advance the RCU grace-period
		 * state machine.
		 */
		smp_mb(); /* see above block comment. */
		rdssp->dynticks++;
		WARN_ON(rdssp->dynticks & 0x1);
	}
}

static void dyntick_save_progress_counter(int cpu)
{
	struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);

	rdssp->dynticks_snap = rdssp->dynticks;
}

static inline int
rcu_try_flip_waitack_needed(int cpu)
{
	long curr;
	long snap;
	struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);

	curr = rdssp->dynticks;
	snap = rdssp->dynticks_snap;
	smp_mb(); /* force ordering with cpu entering/leaving dynticks. */

	/*
	 * If the CPU remained in dynticks mode for the entire time
	 * and didn't take any interrupts, NMIs, SMIs, or whatever,
	 * then it cannot be in the middle of an rcu_read_lock(), so
	 * the next rcu_read_lock() it executes must use the new value
	 * of the counter.  So we can safely pretend that this CPU
	 * already acknowledged the counter.
	 */

	if ((curr == snap) && ((curr & 0x1) == 0))
		return 0;

	/*
	 * If the CPU passed through or entered a dynticks idle phase with
	 * no active irq handlers, then, as above, we can safely pretend
	 * that this CPU already acknowledged the counter.
	 */

	if ((curr - snap) > 2 || (curr & 0x1) == 0)
		return 0;

	/* We need this CPU to explicitly acknowledge the counter flip. */

	return 1;
}

static inline int
rcu_try_flip_waitmb_needed(int cpu)
{
	long curr;
	long snap;
	struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);

	curr = rdssp->dynticks;
	snap = rdssp->dynticks_snap;
	smp_mb(); /* force ordering with cpu entering/leaving dynticks. */

	/*
	 * If the CPU remained in dynticks mode for the entire time
	 * and didn't take any interrupts, NMIs, SMIs, or whatever,
	 * then it cannot have executed an RCU read-side critical section
	 * during that time, so there is no need for it to execute a
	 * memory barrier.
	 */

	if ((curr == snap) && ((curr & 0x1) == 0))
		return 0;

	/*
	 * If the CPU either entered or exited an outermost interrupt,
	 * SMI, NMI, or whatever handler, then we know that it executed
	 * a memory barrier when doing so.  So we don't need another one.
	 */
	if (curr != snap)
		return 0;

	/* We need the CPU to execute a memory barrier. */

	return 1;
}

static void dyntick_save_progress_counter_sched(int cpu)
{
	struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);

	rdssp->sched_dynticks_snap = rdssp->dynticks;
}

static int rcu_qsctr_inc_needed_dyntick(int cpu)
{
	long curr;
	long snap;
	struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);

	curr = rdssp->dynticks;
	snap = rdssp->sched_dynticks_snap;
	smp_mb(); /* force ordering with cpu entering/leaving dynticks. */

	/*
	 * If the CPU remained in dynticks mode for the entire time
	 * and didn't take any interrupts, NMIs, SMIs, or whatever,
	 * then it cannot be in the middle of an rcu_read_lock(), so
	 * the next rcu_read_lock() it executes must use the new value
	 * of the counter.  Therefore, this CPU has been in a quiescent
	 * state the entire time, and we don't need to wait for it.
	 */

	if ((curr == snap) && ((curr & 0x1) == 0))
		return 0;

	/*
	 * If the CPU passed through or entered a dynticks idle phase with
	 * no active irq handlers, then, as above, this CPU has already
	 * passed through a quiescent state.
	 */

	if ((curr - snap) > 2 || (snap & 0x1) == 0)
		return 0;

	/* We need this CPU to go through a quiescent state. */

	return 1;
}

#else /* !CONFIG_NO_HZ */

# define dyntick_save_progress_counter(cpu)		do { } while (0)
# define rcu_try_flip_waitack_needed(cpu)		(1)
# define rcu_try_flip_waitmb_needed(cpu)		(1)

# define dyntick_save_progress_counter_sched(cpu)	do { } while (0)
# define rcu_qsctr_inc_needed_dyntick(cpu)		(1)

#endif /* CONFIG_NO_HZ */

static void save_qsctr_sched(int cpu)
{
	struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);

	rdssp->sched_qs_snap = rdssp->sched_qs;
}

static inline int rcu_qsctr_inc_needed(int cpu)
{
	struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);

	/*
	 * If there has been a quiescent state, no more need to wait
	 * on this CPU.
	 */

	if (rdssp->sched_qs != rdssp->sched_qs_snap) {
		smp_mb(); /* force ordering with cpu entering schedule(). */
		return 0;
	}

	/* We need this CPU to go through a quiescent state. */

	return 1;
}

/*
 * Get here when RCU is idle.  Decide whether we need to
 * move out of idle state, and return non-zero if so.
 * "Straightforward" approach for the moment, might later
 * use callback-list lengths, grace-period duration, or
 * some such to determine when to exit idle state.
 * Might also need a pre-idle test that does not acquire
 * the lock, but let's get the simple case working first...
 */

static int
rcu_try_flip_idle(void)
{
	int cpu;

	RCU_TRACE_ME(rcupreempt_trace_try_flip_i1);
	if (!rcu_pending(smp_processor_id())) {
		RCU_TRACE_ME(rcupreempt_trace_try_flip_ie1);
		return 0;
	}

	/*
	 * Do the flip.
	 */

	RCU_TRACE_ME(rcupreempt_trace_try_flip_g1);
	rcu_ctrlblk.completed++;  /* stands in for rcu_try_flip_g2 */

	/*
	 * Need a memory barrier so that other CPUs see the new
	 * counter value before they see the subsequent change of all
	 * the rcu_flip_flag instances to rcu_flipped.
	 */

	smp_mb();	/* see above block comment. */

	/* Now ask each CPU for acknowledgement of the flip. */

	for_each_cpu_mask_nr(cpu, rcu_cpu_online_map) {
		per_cpu(rcu_flip_flag, cpu) = rcu_flipped;
		dyntick_save_progress_counter(cpu);
	}

	return 1;
}

/*
 * Wait for CPUs to acknowledge the flip.
 */

static int
rcu_try_flip_waitack(void)
{
	int cpu;

	RCU_TRACE_ME(rcupreempt_trace_try_flip_a1);
	for_each_cpu_mask_nr(cpu, rcu_cpu_online_map)
		if (rcu_try_flip_waitack_needed(cpu) &&
		    per_cpu(rcu_flip_flag, cpu) != rcu_flip_seen) {
			RCU_TRACE_ME(rcupreempt_trace_try_flip_ae1);
			return 0;
		}

	/*
	 * Make sure our checks above don't bleed into subsequent
	 * waiting for the sum of the counters to reach zero.
	 */

	smp_mb();	/* see above block comment. */
	RCU_TRACE_ME(rcupreempt_trace_try_flip_a2);
	return 1;
}

/*
 * Wait for collective ``last'' counter to reach zero,
 * then tell all CPUs to do an end-of-grace-period memory barrier.
 */

static int
rcu_try_flip_waitzero(void)
{
	int cpu;
	int lastidx = !(rcu_ctrlblk.completed & 0x1);
	int sum = 0;

	/* Check to see if the sum of the "last" counters is zero. */

	RCU_TRACE_ME(rcupreempt_trace_try_flip_z1);
	for_each_cpu_mask_nr(cpu, rcu_cpu_online_map)
		sum += RCU_DATA_CPU(cpu)->rcu_flipctr[lastidx];
	if (sum != 0) {
		RCU_TRACE_ME(rcupreempt_trace_try_flip_ze1);
		return 0;
	}

	/*
	 * This ensures that the other CPUs see the call for
	 * memory barriers -after- the sum to zero has been
	 * detected here
	 */
	smp_mb();  /*  ^^^^^^^^^^^^ */

	/* Call for a memory barrier from each CPU. */
	for_each_cpu_mask_nr(cpu, rcu_cpu_online_map) {
		per_cpu(rcu_mb_flag, cpu) = rcu_mb_needed;
		dyntick_save_progress_counter(cpu);
	}

	RCU_TRACE_ME(rcupreempt_trace_try_flip_z2);
	return 1;
}

/*
 * Wait for all CPUs to do their end-of-grace-period memory barrier.
 * Return 0 once all CPUs have done so.
 */

static int
rcu_try_flip_waitmb(void)
{
	int cpu;

	RCU_TRACE_ME(rcupreempt_trace_try_flip_m1);
	for_each_cpu_mask_nr(cpu, rcu_cpu_online_map)
		if (rcu_try_flip_waitmb_needed(cpu) &&
		    per_cpu(rcu_mb_flag, cpu) != rcu_mb_done) {
			RCU_TRACE_ME(rcupreempt_trace_try_flip_me1);
			return 0;
		}

	smp_mb(); /* Ensure that the above checks precede any following flip. */
	RCU_TRACE_ME(rcupreempt_trace_try_flip_m2);
	return 1;
}

/*
 * Attempt a single flip of the counters.  Remember, a single flip does
 * -not- constitute a grace period.  Instead, the interval between
 * at least GP_STAGES consecutive flips is a grace period.
 *
 * If anyone is nuts enough to run this CONFIG_PREEMPT_RCU implementation
 * on a large SMP, they might want to use a hierarchical organization of
 * the per-CPU-counter pairs.
 */
static void rcu_try_flip(void)
{
	unsigned long flags;

	RCU_TRACE_ME(rcupreempt_trace_try_flip_1);
	if (unlikely(!spin_trylock_irqsave(&rcu_ctrlblk.fliplock, flags))) {
		RCU_TRACE_ME(rcupreempt_trace_try_flip_e1);
		return;
	}

	/*
	 * Take the next transition(s) through the RCU grace-period
	 * flip-counter state machine.
	 */

	switch (rcu_ctrlblk.rcu_try_flip_state) {
	case rcu_try_flip_idle_state:
		if (rcu_try_flip_idle())
			rcu_ctrlblk.rcu_try_flip_state =
				rcu_try_flip_waitack_state;
		break;
	case rcu_try_flip_waitack_state:
		if (rcu_try_flip_waitack())
			rcu_ctrlblk.rcu_try_flip_state =
				rcu_try_flip_waitzero_state;
		break;
	case rcu_try_flip_waitzero_state:
		if (rcu_try_flip_waitzero())
			rcu_ctrlblk.rcu_try_flip_state =
				rcu_try_flip_waitmb_state;
		break;
	case rcu_try_flip_waitmb_state:
		if (rcu_try_flip_waitmb())
			rcu_ctrlblk.rcu_try_flip_state =
				rcu_try_flip_idle_state;
	}
	spin_unlock_irqrestore(&rcu_ctrlblk.fliplock, flags);
}

/*
 * Check to see if this CPU needs to do a memory barrier in order to
 * ensure that any prior RCU read-side critical sections have committed
 * their counter manipulations and critical-section memory references
 * before declaring the grace period to be completed.
 */
static void rcu_check_mb(int cpu)
{
	if (per_cpu(rcu_mb_flag, cpu) == rcu_mb_needed) {
		smp_mb();  /* Ensure RCU read-side accesses are visible. */
		per_cpu(rcu_mb_flag, cpu) = rcu_mb_done;
	}
}

void rcu_check_callbacks(int cpu, int user)
{
	unsigned long flags;
	struct rcu_data *rdp = RCU_DATA_CPU(cpu);

	/*
	 * If this CPU took its interrupt from user mode or from the
	 * idle loop, and this is not a nested interrupt, then
	 * this CPU has to have exited all prior preept-disable
	 * sections of code.  So increment the counter to note this.
	 *
	 * The memory barrier is needed to handle the case where
	 * writes from a preempt-disable section of code get reordered
	 * into schedule() by this CPU's write buffer.  So the memory
	 * barrier makes sure that the rcu_qsctr_inc() is seen by other
	 * CPUs to happen after any such write.
	 */

	if (user ||
	    (idle_cpu(cpu) && !in_softirq() &&
	     hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
		smp_mb();	/* Guard against aggressive schedule(). */
	     	rcu_qsctr_inc(cpu);
	}

	rcu_check_mb(cpu);
	if (rcu_ctrlblk.completed == rdp->completed)
		rcu_try_flip();
	spin_lock_irqsave(&rdp->lock, flags);
	RCU_TRACE_RDP(rcupreempt_trace_check_callbacks, rdp);
	__rcu_advance_callbacks(rdp);
	if (rdp->donelist == NULL) {
		spin_unlock_irqrestore(&rdp->lock, flags);
	} else {
		spin_unlock_irqrestore(&rdp->lock, flags);
		raise_softirq(RCU_SOFTIRQ);
	}
}

/*
 * Needed by dynticks, to make sure all RCU processing has finished
 * when we go idle:
 */
void rcu_advance_callbacks(int cpu, int user)
{
	unsigned long flags;
	struct rcu_data *rdp = RCU_DATA_CPU(cpu);

	if (rcu_ctrlblk.completed == rdp->completed) {
		rcu_try_flip();
		if (rcu_ctrlblk.completed == rdp->completed)
			return;
	}
	spin_lock_irqsave(&rdp->lock, flags);
	RCU_TRACE_RDP(rcupreempt_trace_check_callbacks, rdp);
	__rcu_advance_callbacks(rdp);
	spin_unlock_irqrestore(&rdp->lock, flags);
}

#ifdef CONFIG_HOTPLUG_CPU
#define rcu_offline_cpu_enqueue(srclist, srctail, dstlist, dsttail) do { \
		*dsttail = srclist; \
		if (srclist != NULL) { \
			dsttail = srctail; \
			srclist = NULL; \
			srctail = &srclist;\
		} \
	} while (0)

void rcu_offline_cpu(int cpu)
{
	int i;
	struct rcu_head *list = NULL;
	unsigned long flags;
	struct rcu_data *rdp = RCU_DATA_CPU(cpu);
	struct rcu_head *schedlist = NULL;
	struct rcu_head **schedtail = &schedlist;
	struct rcu_head **tail = &list;

	/*
	 * Remove all callbacks from the newly dead CPU, retaining order.
	 * Otherwise rcu_barrier() will fail
	 */

	spin_lock_irqsave(&rdp->lock, flags);
	rcu_offline_cpu_enqueue(rdp->donelist, rdp->donetail, list, tail);
	for (i = GP_STAGES - 1; i >= 0; i--)
		rcu_offline_cpu_enqueue(rdp->waitlist[i], rdp->waittail[i],
						list, tail);
	rcu_offline_cpu_enqueue(rdp->nextlist, rdp->nexttail, list, tail);
	rcu_offline_cpu_enqueue(rdp->waitschedlist, rdp->waitschedtail,
				schedlist, schedtail);
	rcu_offline_cpu_enqueue(rdp->nextschedlist, rdp->nextschedtail,
				schedlist, schedtail);
	rdp->rcu_sched_sleeping = 0;
	spin_unlock_irqrestore(&rdp->lock, flags);
	rdp->waitlistcount = 0;

	/* Disengage the newly dead CPU from the grace-period computation. */

	spin_lock_irqsave(&rcu_ctrlblk.fliplock, flags);
	rcu_check_mb(cpu);
	if (per_cpu(rcu_flip_flag, cpu) == rcu_flipped) {
		smp_mb();  /* Subsequent counter accesses must see new value */
		per_cpu(rcu_flip_flag, cpu) = rcu_flip_seen;
		smp_mb();  /* Subsequent RCU read-side critical sections */
			   /*  seen -after- acknowledgement. */
	}

	RCU_DATA_ME()->rcu_flipctr[0] += RCU_DATA_CPU(cpu)->rcu_flipctr[0];
	RCU_DATA_ME()->rcu_flipctr[1] += RCU_DATA_CPU(cpu)->rcu_flipctr[1];

	RCU_DATA_CPU(cpu)->rcu_flipctr[0] = 0;
	RCU_DATA_CPU(cpu)->rcu_flipctr[1] = 0;

	cpu_clear(cpu, rcu_cpu_online_map);

	spin_unlock_irqrestore(&rcu_ctrlblk.fliplock, flags);

	/*
	 * Place the removed callbacks on the current CPU's queue.
	 * Make them all start a new grace period: simple approach,
	 * in theory could starve a given set of callbacks, but
	 * you would need to be doing some serious CPU hotplugging
	 * to make this happen.  If this becomes a problem, adding
	 * a synchronize_rcu() to the hotplug path would be a simple
	 * fix.
	 */

	local_irq_save(flags);  /* disable preempt till we know what lock. */
	rdp = RCU_DATA_ME();
	spin_lock(&rdp->lock);
	*rdp->nexttail = list;
	if (list)
		rdp->nexttail = tail;
	*rdp->nextschedtail = schedlist;
	if (schedlist)
		rdp->nextschedtail = schedtail;
	spin_unlock_irqrestore(&rdp->lock, flags);
}

#else /* #ifdef CONFIG_HOTPLUG_CPU */

void rcu_offline_cpu(int cpu)
{
}

#endif /* #else #ifdef CONFIG_HOTPLUG_CPU */

void __cpuinit rcu_online_cpu(int cpu)
{
	unsigned long flags;
	struct rcu_data *rdp;

	spin_lock_irqsave(&rcu_ctrlblk.fliplock, flags);
	cpu_set(cpu, rcu_cpu_online_map);
	spin_unlock_irqrestore(&rcu_ctrlblk.fliplock, flags);

	/*
	 * The rcu_sched grace-period processing might have bypassed
	 * this CPU, given that it was not in the rcu_cpu_online_map
	 * when the grace-period scan started.  This means that the
	 * grace-period task might sleep.  So make sure that if this
	 * should happen, the first callback posted to this CPU will
	 * wake up the grace-period task if need be.
	 */

	rdp = RCU_DATA_CPU(cpu);
	spin_lock_irqsave(&rdp->lock, flags);
	rdp->rcu_sched_sleeping = 1;
	spin_unlock_irqrestore(&rdp->lock, flags);
}

static void rcu_process_callbacks(struct softirq_action *unused)
{
	unsigned long flags;
	struct rcu_head *next, *list;
	struct rcu_data *rdp;

	local_irq_save(flags);
	rdp = RCU_DATA_ME();
	spin_lock(&rdp->lock);
	list = rdp->donelist;
	if (list == NULL) {
		spin_unlock_irqrestore(&rdp->lock, flags);
		return;
	}
	rdp->donelist = NULL;
	rdp->donetail = &rdp->donelist;
	RCU_TRACE_RDP(rcupreempt_trace_done_remove, rdp);
	spin_unlock_irqrestore(&rdp->lock, flags);
	while (list) {
		next = list->next;
		list->func(list);
		list = next;
		RCU_TRACE_ME(rcupreempt_trace_invoke);
	}
}

void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
{
	unsigned long flags;
	struct rcu_data *rdp;

	head->func = func;
	head->next = NULL;
	local_irq_save(flags);
	rdp = RCU_DATA_ME();
	spin_lock(&rdp->lock);
	__rcu_advance_callbacks(rdp);
	*rdp->nexttail = head;
	rdp->nexttail = &head->next;
	RCU_TRACE_RDP(rcupreempt_trace_next_add, rdp);
	spin_unlock_irqrestore(&rdp->lock, flags);
}
EXPORT_SYMBOL_GPL(call_rcu);

void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
{
	unsigned long flags;
	struct rcu_data *rdp;
	int wake_gp = 0;

	head->func = func;
	head->next = NULL;
	local_irq_save(flags);
	rdp = RCU_DATA_ME();
	spin_lock(&rdp->lock);
	*rdp->nextschedtail = head;
	rdp->nextschedtail = &head->next;
	if (rdp->rcu_sched_sleeping) {

		/* Grace-period processing might be sleeping... */

		rdp->rcu_sched_sleeping = 0;
		wake_gp = 1;
	}
	spin_unlock_irqrestore(&rdp->lock, flags);
	if (wake_gp) {

		/* Wake up grace-period processing, unless someone beat us. */

		spin_lock_irqsave(&rcu_ctrlblk.schedlock, flags);
		if (rcu_ctrlblk.sched_sleep != rcu_sched_sleeping)
			wake_gp = 0;
		rcu_ctrlblk.sched_sleep = rcu_sched_not_sleeping;
		spin_unlock_irqrestore(&rcu_ctrlblk.schedlock, flags);
		if (wake_gp)
			wake_up_interruptible(&rcu_ctrlblk.sched_wq);
	}
}
EXPORT_SYMBOL_GPL(call_rcu_sched);

/*
 * Wait until all currently running preempt_disable() code segments
 * (including hardware-irq-disable segments) complete.  Note that
 * in -rt this does -not- necessarily result in all currently executing
 * interrupt -handlers- having completed.
 */
synchronize_rcu_xxx(__synchronize_sched, call_rcu_sched)
EXPORT_SYMBOL_GPL(__synchronize_sched);

/*
 * kthread function that manages call_rcu_sched grace periods.
 */
static int rcu_sched_grace_period(void *arg)
{
	int couldsleep;		/* might sleep after current pass. */
	int couldsleepnext = 0; /* might sleep after next pass. */
	int cpu;
	unsigned long flags;
	struct rcu_data *rdp;
	int ret;

	/*
	 * Each pass through the following loop handles one
	 * rcu_sched grace period cycle.
	 */
	do {
		/* Save each CPU's current state. */

		for_each_online_cpu(cpu) {
			dyntick_save_progress_counter_sched(cpu);
			save_qsctr_sched(cpu);
		}

		/*
		 * Sleep for about an RCU grace-period's worth to
		 * allow better batching and to consume less CPU.
		 */
		schedule_timeout_interruptible(RCU_SCHED_BATCH_TIME);

		/*
		 * If there was nothing to do last time, prepare to
		 * sleep at the end of the current grace period cycle.
		 */
		couldsleep = couldsleepnext;
		couldsleepnext = 1;
		if (couldsleep) {
			spin_lock_irqsave(&rcu_ctrlblk.schedlock, flags);
			rcu_ctrlblk.sched_sleep = rcu_sched_sleep_prep;
			spin_unlock_irqrestore(&rcu_ctrlblk.schedlock, flags);
		}

		/*
		 * Wait on each CPU in turn to have either visited
		 * a quiescent state or been in dynticks-idle mode.
		 */
		for_each_online_cpu(cpu) {
			while (rcu_qsctr_inc_needed(cpu) &&
			       rcu_qsctr_inc_needed_dyntick(cpu)) {
				/* resched_cpu(cpu); @@@ */
				schedule_timeout_interruptible(1);
			}
		}

		/* Advance callbacks for each CPU.  */

		for_each_online_cpu(cpu) {

			rdp = RCU_DATA_CPU(cpu);
			spin_lock_irqsave(&rdp->lock, flags);

			/*
			 * We are running on this CPU irq-disabled, so no
			 * CPU can go offline until we re-enable irqs.
			 * The current CPU might have already gone
			 * offline (between the for_each_offline_cpu and
			 * the spin_lock_irqsave), but in that case all its
			 * callback lists will be empty, so no harm done.
			 *
			 * Advance the callbacks!  We share normal RCU's
			 * donelist, since callbacks are invoked the
			 * same way in either case.
			 */
			if (rdp->waitschedlist != NULL) {
				*rdp->donetail = rdp->waitschedlist;
				rdp->donetail = rdp->waitschedtail;

				/*
				 * Next rcu_check_callbacks() will
				 * do the required raise_softirq().
				 */
			}
			if (rdp->nextschedlist != NULL) {
				rdp->waitschedlist = rdp->nextschedlist;
				rdp->waitschedtail = rdp->nextschedtail;
				couldsleep = 0;
				couldsleepnext = 0;
			} else {
				rdp->waitschedlist = NULL;
				rdp->waitschedtail = &rdp->waitschedlist;
			}
			rdp->nextschedlist = NULL;
			rdp->nextschedtail = &rdp->nextschedlist;

			/* Mark sleep intention. */

			rdp->rcu_sched_sleeping = couldsleep;

			spin_unlock_irqrestore(&rdp->lock, flags);
		}

		/* If we saw callbacks on the last scan, go deal with them. */

		if (!couldsleep)
			continue;

		/* Attempt to block... */

		spin_lock_irqsave(&rcu_ctrlblk.schedlock, flags);
		if (rcu_ctrlblk.sched_sleep != rcu_sched_sleep_prep) {

			/*
			 * Someone posted a callback after we scanned.
			 * Go take care of it.
			 */
			spin_unlock_irqrestore(&rcu_ctrlblk.schedlock, flags);
			couldsleepnext = 0;
			continue;
		}

		/* Block until the next person posts a callback. */

		rcu_ctrlblk.sched_sleep = rcu_sched_sleeping;
		spin_unlock_irqrestore(&rcu_ctrlblk.schedlock, flags);
		ret = 0;
		__wait_event_interruptible(rcu_ctrlblk.sched_wq,
			rcu_ctrlblk.sched_sleep != rcu_sched_sleeping,
			ret);

		/*
		 * Signals would prevent us from sleeping, and we cannot
		 * do much with them in any case.  So flush them.
		 */
		if (ret)
			flush_signals(current);
		couldsleepnext = 0;

	} while (!kthread_should_stop());

	return (0);
}

/*
 * Check to see if any future RCU-related work will need to be done
 * by the current CPU, even if none need be done immediately, returning
 * 1 if so.  Assumes that notifiers would take care of handling any
 * outstanding requests from the RCU core.
 *
 * This function is part of the RCU implementation; it is -not-
 * an exported member of the RCU API.
 */
int rcu_needs_cpu(int cpu)
{
	struct rcu_data *rdp = RCU_DATA_CPU(cpu);

	return (rdp->donelist != NULL ||
		!!rdp->waitlistcount ||
		rdp->nextlist != NULL ||
		rdp->nextschedlist != NULL ||
		rdp->waitschedlist != NULL);
}

int rcu_pending(int cpu)
{
	struct rcu_data *rdp = RCU_DATA_CPU(cpu);

	/* The CPU has at least one callback queued somewhere. */

	if (rdp->donelist != NULL ||
	    !!rdp->waitlistcount ||
	    rdp->nextlist != NULL ||
	    rdp->nextschedlist != NULL ||
	    rdp->waitschedlist != NULL)
		return 1;

	/* The RCU core needs an acknowledgement from this CPU. */

	if ((per_cpu(rcu_flip_flag, cpu) == rcu_flipped) ||
	    (per_cpu(rcu_mb_flag, cpu) == rcu_mb_needed))
		return 1;

	/* This CPU has fallen behind the global grace-period number. */

	if (rdp->completed != rcu_ctrlblk.completed)
		return 1;

	/* Nothing needed from this CPU. */

	return 0;
}

static int __cpuinit rcu_cpu_notify(struct notifier_block *self,
				unsigned long action, void *hcpu)
{
	long cpu = (long)hcpu;

	switch (action) {
	case CPU_UP_PREPARE:
	case CPU_UP_PREPARE_FROZEN:
		rcu_online_cpu(cpu);
		break;
	case CPU_UP_CANCELED:
	case CPU_UP_CANCELED_FROZEN:
	case CPU_DEAD:
	case CPU_DEAD_FROZEN:
		rcu_offline_cpu(cpu);
		break;
	default:
		break;
	}
	return NOTIFY_OK;
}

static struct notifier_block __cpuinitdata rcu_nb = {
	.notifier_call = rcu_cpu_notify,
};

void __init __rcu_init(void)
{
	int cpu;
	int i;
	struct rcu_data *rdp;

	printk(KERN_NOTICE "Preemptible RCU implementation.\n");
	for_each_possible_cpu(cpu) {
		rdp = RCU_DATA_CPU(cpu);
		spin_lock_init(&rdp->lock);
		rdp->completed = 0;
		rdp->waitlistcount = 0;
		rdp->nextlist = NULL;
		rdp->nexttail = &rdp->nextlist;
		for (i = 0; i < GP_STAGES; i++) {
			rdp->waitlist[i] = NULL;
			rdp->waittail[i] = &rdp->waitlist[i];
		}
		rdp->donelist = NULL;
		rdp->donetail = &rdp->donelist;
		rdp->rcu_flipctr[0] = 0;
		rdp->rcu_flipctr[1] = 0;
		rdp->nextschedlist = NULL;
		rdp->nextschedtail = &rdp->nextschedlist;
		rdp->waitschedlist = NULL;
		rdp->waitschedtail = &rdp->waitschedlist;
		rdp->rcu_sched_sleeping = 0;
	}
	register_cpu_notifier(&rcu_nb);

	/*
	 * We don't need protection against CPU-Hotplug here
	 * since
	 * a) If a CPU comes online while we are iterating over the
	 *    cpu_online_map below, we would only end up making a
	 *    duplicate call to rcu_online_cpu() which sets the corresponding
	 *    CPU's mask in the rcu_cpu_online_map.
	 *
	 * b) A CPU cannot go offline at this point in time since the user
	 *    does not have access to the sysfs interface, nor do we
	 *    suspend the system.
	 */
	for_each_online_cpu(cpu)
		rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE,	(void *)(long) cpu);

	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
}

/*
 * Late-boot-time RCU initialization that must wait until after scheduler
 * has been initialized.
 */
void __init rcu_init_sched(void)
{
	rcu_sched_grace_period_task = kthread_run(rcu_sched_grace_period,
						  NULL,
						  "rcu_sched_grace_period");
	WARN_ON(IS_ERR(rcu_sched_grace_period_task));
}

#ifdef CONFIG_RCU_TRACE
long *rcupreempt_flipctr(int cpu)
{
	return &RCU_DATA_CPU(cpu)->rcu_flipctr[0];
}
EXPORT_SYMBOL_GPL(rcupreempt_flipctr);

int rcupreempt_flip_flag(int cpu)
{
	return per_cpu(rcu_flip_flag, cpu);
}
EXPORT_SYMBOL_GPL(rcupreempt_flip_flag);

int rcupreempt_mb_flag(int cpu)
{
	return per_cpu(rcu_mb_flag, cpu);
}
EXPORT_SYMBOL_GPL(rcupreempt_mb_flag);

char *rcupreempt_try_flip_state_name(void)
{
	return rcu_try_flip_state_names[rcu_ctrlblk.rcu_try_flip_state];
}
EXPORT_SYMBOL_GPL(rcupreempt_try_flip_state_name);

struct rcupreempt_trace *rcupreempt_trace_cpu(int cpu)
{
	struct rcu_data *rdp = RCU_DATA_CPU(cpu);

	return &rdp->trace;
}
EXPORT_SYMBOL_GPL(rcupreempt_trace_cpu);

#endif /* #ifdef RCU_TRACE */