summaryrefslogtreecommitdiff
path: root/arch/powerpc/kernel/perf_counter.c
blob: 7ceefaf3a7f5e51af1fb44dff457117d7034f4f4 (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
/*
 * Performance counter support - powerpc architecture code
 *
 * Copyright 2008-2009 Paul Mackerras, IBM Corporation.
 *
 * 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.
 */
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/perf_counter.h>
#include <linux/percpu.h>
#include <linux/hardirq.h>
#include <asm/reg.h>
#include <asm/pmc.h>
#include <asm/machdep.h>
#include <asm/firmware.h>
#include <asm/ptrace.h>

struct cpu_hw_counters {
	int n_counters;
	int n_percpu;
	int disabled;
	int n_added;
	int n_limited;
	u8  pmcs_enabled;
	struct perf_counter *counter[MAX_HWCOUNTERS];
	u64 events[MAX_HWCOUNTERS];
	unsigned int flags[MAX_HWCOUNTERS];
	unsigned long mmcr[3];
	struct perf_counter *limited_counter[MAX_LIMITED_HWCOUNTERS];
	u8  limited_hwidx[MAX_LIMITED_HWCOUNTERS];
	u64 alternatives[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES];
	unsigned long amasks[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES];
	unsigned long avalues[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES];
};
DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters);

struct power_pmu *ppmu;

/*
 * Normally, to ignore kernel events we set the FCS (freeze counters
 * in supervisor mode) bit in MMCR0, but if the kernel runs with the
 * hypervisor bit set in the MSR, or if we are running on a processor
 * where the hypervisor bit is forced to 1 (as on Apple G5 processors),
 * then we need to use the FCHV bit to ignore kernel events.
 */
static unsigned int freeze_counters_kernel = MMCR0_FCS;

/*
 * 32-bit doesn't have MMCRA but does have an MMCR2,
 * and a few other names are different.
 */
#ifdef CONFIG_PPC32

#define MMCR0_FCHV		0
#define MMCR0_PMCjCE		MMCR0_PMCnCE

#define SPRN_MMCRA		SPRN_MMCR2
#define MMCRA_SAMPLE_ENABLE	0

static inline unsigned long perf_ip_adjust(struct pt_regs *regs)
{
	return 0;
}
static inline void perf_get_data_addr(struct pt_regs *regs, u64 *addrp) { }
static inline u32 perf_get_misc_flags(struct pt_regs *regs)
{
	return 0;
}
static inline void perf_read_regs(struct pt_regs *regs) { }
static inline int perf_intr_is_nmi(struct pt_regs *regs)
{
	return 0;
}

#endif /* CONFIG_PPC32 */

/*
 * Things that are specific to 64-bit implementations.
 */
#ifdef CONFIG_PPC64

static inline unsigned long perf_ip_adjust(struct pt_regs *regs)
{
	unsigned long mmcra = regs->dsisr;

	if ((mmcra & MMCRA_SAMPLE_ENABLE) && !(ppmu->flags & PPMU_ALT_SIPR)) {
		unsigned long slot = (mmcra & MMCRA_SLOT) >> MMCRA_SLOT_SHIFT;
		if (slot > 1)
			return 4 * (slot - 1);
	}
	return 0;
}

/*
 * The user wants a data address recorded.
 * If we're not doing instruction sampling, give them the SDAR
 * (sampled data address).  If we are doing instruction sampling, then
 * only give them the SDAR if it corresponds to the instruction
 * pointed to by SIAR; this is indicated by the [POWER6_]MMCRA_SDSYNC
 * bit in MMCRA.
 */
static inline void perf_get_data_addr(struct pt_regs *regs, u64 *addrp)
{
	unsigned long mmcra = regs->dsisr;
	unsigned long sdsync = (ppmu->flags & PPMU_ALT_SIPR) ?
		POWER6_MMCRA_SDSYNC : MMCRA_SDSYNC;

	if (!(mmcra & MMCRA_SAMPLE_ENABLE) || (mmcra & sdsync))
		*addrp = mfspr(SPRN_SDAR);
}

static inline u32 perf_get_misc_flags(struct pt_regs *regs)
{
	unsigned long mmcra = regs->dsisr;

	if (TRAP(regs) != 0xf00)
		return 0;	/* not a PMU interrupt */

	if (ppmu->flags & PPMU_ALT_SIPR) {
		if (mmcra & POWER6_MMCRA_SIHV)
			return PERF_EVENT_MISC_HYPERVISOR;
		return (mmcra & POWER6_MMCRA_SIPR) ?
			PERF_EVENT_MISC_USER : PERF_EVENT_MISC_KERNEL;
	}
	if (mmcra & MMCRA_SIHV)
		return PERF_EVENT_MISC_HYPERVISOR;
	return (mmcra & MMCRA_SIPR) ? PERF_EVENT_MISC_USER :
		PERF_EVENT_MISC_KERNEL;
}

/*
 * Overload regs->dsisr to store MMCRA so we only need to read it once
 * on each interrupt.
 */
static inline void perf_read_regs(struct pt_regs *regs)
{
	regs->dsisr = mfspr(SPRN_MMCRA);
}

/*
 * If interrupts were soft-disabled when a PMU interrupt occurs, treat
 * it as an NMI.
 */
static inline int perf_intr_is_nmi(struct pt_regs *regs)
{
	return !regs->softe;
}

#endif /* CONFIG_PPC64 */

static void perf_counter_interrupt(struct pt_regs *regs);

void perf_counter_print_debug(void)
{
}

/*
 * Read one performance monitor counter (PMC).
 */
static unsigned long read_pmc(int idx)
{
	unsigned long val;

	switch (idx) {
	case 1:
		val = mfspr(SPRN_PMC1);
		break;
	case 2:
		val = mfspr(SPRN_PMC2);
		break;
	case 3:
		val = mfspr(SPRN_PMC3);
		break;
	case 4:
		val = mfspr(SPRN_PMC4);
		break;
	case 5:
		val = mfspr(SPRN_PMC5);
		break;
	case 6:
		val = mfspr(SPRN_PMC6);
		break;
#ifdef CONFIG_PPC64
	case 7:
		val = mfspr(SPRN_PMC7);
		break;
	case 8:
		val = mfspr(SPRN_PMC8);
		break;
#endif /* CONFIG_PPC64 */
	default:
		printk(KERN_ERR "oops trying to read PMC%d\n", idx);
		val = 0;
	}
	return val;
}

/*
 * Write one PMC.
 */
static void write_pmc(int idx, unsigned long val)
{
	switch (idx) {
	case 1:
		mtspr(SPRN_PMC1, val);
		break;
	case 2:
		mtspr(SPRN_PMC2, val);
		break;
	case 3:
		mtspr(SPRN_PMC3, val);
		break;
	case 4:
		mtspr(SPRN_PMC4, val);
		break;
	case 5:
		mtspr(SPRN_PMC5, val);
		break;
	case 6:
		mtspr(SPRN_PMC6, val);
		break;
#ifdef CONFIG_PPC64
	case 7:
		mtspr(SPRN_PMC7, val);
		break;
	case 8:
		mtspr(SPRN_PMC8, val);
		break;
#endif /* CONFIG_PPC64 */
	default:
		printk(KERN_ERR "oops trying to write PMC%d\n", idx);
	}
}

/*
 * Check if a set of events can all go on the PMU at once.
 * If they can't, this will look at alternative codes for the events
 * and see if any combination of alternative codes is feasible.
 * The feasible set is returned in event[].
 */
static int power_check_constraints(struct cpu_hw_counters *cpuhw,
				   u64 event[], unsigned int cflags[],
				   int n_ev)
{
	unsigned long mask, value, nv;
	unsigned long smasks[MAX_HWCOUNTERS], svalues[MAX_HWCOUNTERS];
	int n_alt[MAX_HWCOUNTERS], choice[MAX_HWCOUNTERS];
	int i, j;
	unsigned long addf = ppmu->add_fields;
	unsigned long tadd = ppmu->test_adder;

	if (n_ev > ppmu->n_counter)
		return -1;

	/* First see if the events will go on as-is */
	for (i = 0; i < n_ev; ++i) {
		if ((cflags[i] & PPMU_LIMITED_PMC_REQD)
		    && !ppmu->limited_pmc_event(event[i])) {
			ppmu->get_alternatives(event[i], cflags[i],
					       cpuhw->alternatives[i]);
			event[i] = cpuhw->alternatives[i][0];
		}
		if (ppmu->get_constraint(event[i], &cpuhw->amasks[i][0],
					 &cpuhw->avalues[i][0]))
			return -1;
	}
	value = mask = 0;
	for (i = 0; i < n_ev; ++i) {
		nv = (value | cpuhw->avalues[i][0]) +
			(value & cpuhw->avalues[i][0] & addf);
		if ((((nv + tadd) ^ value) & mask) != 0 ||
		    (((nv + tadd) ^ cpuhw->avalues[i][0]) &
		     cpuhw->amasks[i][0]) != 0)
			break;
		value = nv;
		mask |= cpuhw->amasks[i][0];
	}
	if (i == n_ev)
		return 0;	/* all OK */

	/* doesn't work, gather alternatives... */
	if (!ppmu->get_alternatives)
		return -1;
	for (i = 0; i < n_ev; ++i) {
		choice[i] = 0;
		n_alt[i] = ppmu->get_alternatives(event[i], cflags[i],
						  cpuhw->alternatives[i]);
		for (j = 1; j < n_alt[i]; ++j)
			ppmu->get_constraint(cpuhw->alternatives[i][j],
					     &cpuhw->amasks[i][j],
					     &cpuhw->avalues[i][j]);
	}

	/* enumerate all possibilities and see if any will work */
	i = 0;
	j = -1;
	value = mask = nv = 0;
	while (i < n_ev) {
		if (j >= 0) {
			/* we're backtracking, restore context */
			value = svalues[i];
			mask = smasks[i];
			j = choice[i];
		}
		/*
		 * See if any alternative k for event i,
		 * where k > j, will satisfy the constraints.
		 */
		while (++j < n_alt[i]) {
			nv = (value | cpuhw->avalues[i][j]) +
				(value & cpuhw->avalues[i][j] & addf);
			if ((((nv + tadd) ^ value) & mask) == 0 &&
			    (((nv + tadd) ^ cpuhw->avalues[i][j])
			     & cpuhw->amasks[i][j]) == 0)
				break;
		}
		if (j >= n_alt[i]) {
			/*
			 * No feasible alternative, backtrack
			 * to event i-1 and continue enumerating its
			 * alternatives from where we got up to.
			 */
			if (--i < 0)
				return -1;
		} else {
			/*
			 * Found a feasible alternative for event i,
			 * remember where we got up to with this event,
			 * go on to the next event, and start with
			 * the first alternative for it.
			 */
			choice[i] = j;
			svalues[i] = value;
			smasks[i] = mask;
			value = nv;
			mask |= cpuhw->amasks[i][j];
			++i;
			j = -1;
		}
	}

	/* OK, we have a feasible combination, tell the caller the solution */
	for (i = 0; i < n_ev; ++i)
		event[i] = cpuhw->alternatives[i][choice[i]];
	return 0;
}

/*
 * Check if newly-added counters have consistent settings for
 * exclude_{user,kernel,hv} with each other and any previously
 * added counters.
 */
static int check_excludes(struct perf_counter **ctrs, unsigned int cflags[],
			  int n_prev, int n_new)
{
	int eu = 0, ek = 0, eh = 0;
	int i, n, first;
	struct perf_counter *counter;

	n = n_prev + n_new;
	if (n <= 1)
		return 0;

	first = 1;
	for (i = 0; i < n; ++i) {
		if (cflags[i] & PPMU_LIMITED_PMC_OK) {
			cflags[i] &= ~PPMU_LIMITED_PMC_REQD;
			continue;
		}
		counter = ctrs[i];
		if (first) {
			eu = counter->attr.exclude_user;
			ek = counter->attr.exclude_kernel;
			eh = counter->attr.exclude_hv;
			first = 0;
		} else if (counter->attr.exclude_user != eu ||
			   counter->attr.exclude_kernel != ek ||
			   counter->attr.exclude_hv != eh) {
			return -EAGAIN;
		}
	}

	if (eu || ek || eh)
		for (i = 0; i < n; ++i)
			if (cflags[i] & PPMU_LIMITED_PMC_OK)
				cflags[i] |= PPMU_LIMITED_PMC_REQD;

	return 0;
}

static void power_pmu_read(struct perf_counter *counter)
{
	s64 val, delta, prev;

	if (!counter->hw.idx)
		return;
	/*
	 * Performance monitor interrupts come even when interrupts
	 * are soft-disabled, as long as interrupts are hard-enabled.
	 * Therefore we treat them like NMIs.
	 */
	do {
		prev = atomic64_read(&counter->hw.prev_count);
		barrier();
		val = read_pmc(counter->hw.idx);
	} while (atomic64_cmpxchg(&counter->hw.prev_count, prev, val) != prev);

	/* The counters are only 32 bits wide */
	delta = (val - prev) & 0xfffffffful;
	atomic64_add(delta, &counter->count);
	atomic64_sub(delta, &counter->hw.period_left);
}

/*
 * On some machines, PMC5 and PMC6 can't be written, don't respect
 * the freeze conditions, and don't generate interrupts.  This tells
 * us if `counter' is using such a PMC.
 */
static int is_limited_pmc(int pmcnum)
{
	return (ppmu->flags & PPMU_LIMITED_PMC5_6)
		&& (pmcnum == 5 || pmcnum == 6);
}

static void freeze_limited_counters(struct cpu_hw_counters *cpuhw,
				    unsigned long pmc5, unsigned long pmc6)
{
	struct perf_counter *counter;
	u64 val, prev, delta;
	int i;

	for (i = 0; i < cpuhw->n_limited; ++i) {
		counter = cpuhw->limited_counter[i];
		if (!counter->hw.idx)
			continue;
		val = (counter->hw.idx == 5) ? pmc5 : pmc6;
		prev = atomic64_read(&counter->hw.prev_count);
		counter->hw.idx = 0;
		delta = (val - prev) & 0xfffffffful;
		atomic64_add(delta, &counter->count);
	}
}

static void thaw_limited_counters(struct cpu_hw_counters *cpuhw,
				  unsigned long pmc5, unsigned long pmc6)
{
	struct perf_counter *counter;
	u64 val;
	int i;

	for (i = 0; i < cpuhw->n_limited; ++i) {
		counter = cpuhw->limited_counter[i];
		counter->hw.idx = cpuhw->limited_hwidx[i];
		val = (counter->hw.idx == 5) ? pmc5 : pmc6;
		atomic64_set(&counter->hw.prev_count, val);
		perf_counter_update_userpage(counter);
	}
}

/*
 * Since limited counters don't respect the freeze conditions, we
 * have to read them immediately after freezing or unfreezing the
 * other counters.  We try to keep the values from the limited
 * counters as consistent as possible by keeping the delay (in
 * cycles and instructions) between freezing/unfreezing and reading
 * the limited counters as small and consistent as possible.
 * Therefore, if any limited counters are in use, we read them
 * both, and always in the same order, to minimize variability,
 * and do it inside the same asm that writes MMCR0.
 */
static void write_mmcr0(struct cpu_hw_counters *cpuhw, unsigned long mmcr0)
{
	unsigned long pmc5, pmc6;

	if (!cpuhw->n_limited) {
		mtspr(SPRN_MMCR0, mmcr0);
		return;
	}

	/*
	 * Write MMCR0, then read PMC5 and PMC6 immediately.
	 * To ensure we don't get a performance monitor interrupt
	 * between writing MMCR0 and freezing/thawing the limited
	 * counters, we first write MMCR0 with the counter overflow
	 * interrupt enable bits turned off.
	 */
	asm volatile("mtspr %3,%2; mfspr %0,%4; mfspr %1,%5"
		     : "=&r" (pmc5), "=&r" (pmc6)
		     : "r" (mmcr0 & ~(MMCR0_PMC1CE | MMCR0_PMCjCE)),
		       "i" (SPRN_MMCR0),
		       "i" (SPRN_PMC5), "i" (SPRN_PMC6));

	if (mmcr0 & MMCR0_FC)
		freeze_limited_counters(cpuhw, pmc5, pmc6);
	else
		thaw_limited_counters(cpuhw, pmc5, pmc6);

	/*
	 * Write the full MMCR0 including the counter overflow interrupt
	 * enable bits, if necessary.
	 */
	if (mmcr0 & (MMCR0_PMC1CE | MMCR0_PMCjCE))
		mtspr(SPRN_MMCR0, mmcr0);
}

/*
 * Disable all counters to prevent PMU interrupts and to allow
 * counters to be added or removed.
 */
void hw_perf_disable(void)
{
	struct cpu_hw_counters *cpuhw;
	unsigned long flags;

	if (!ppmu)
		return;
	local_irq_save(flags);
	cpuhw = &__get_cpu_var(cpu_hw_counters);

	if (!cpuhw->disabled) {
		cpuhw->disabled = 1;
		cpuhw->n_added = 0;

		/*
		 * Check if we ever enabled the PMU on this cpu.
		 */
		if (!cpuhw->pmcs_enabled) {
			ppc_enable_pmcs();
			cpuhw->pmcs_enabled = 1;
		}

		/*
		 * Disable instruction sampling if it was enabled
		 */
		if (cpuhw->mmcr[2] & MMCRA_SAMPLE_ENABLE) {
			mtspr(SPRN_MMCRA,
			      cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE);
			mb();
		}

		/*
		 * Set the 'freeze counters' bit.
		 * The barrier is to make sure the mtspr has been
		 * executed and the PMU has frozen the counters
		 * before we return.
		 */
		write_mmcr0(cpuhw, mfspr(SPRN_MMCR0) | MMCR0_FC);
		mb();
	}
	local_irq_restore(flags);
}

/*
 * Re-enable all counters if disable == 0.
 * If we were previously disabled and counters were added, then
 * put the new config on the PMU.
 */
void hw_perf_enable(void)
{
	struct perf_counter *counter;
	struct cpu_hw_counters *cpuhw;
	unsigned long flags;
	long i;
	unsigned long val;
	s64 left;
	unsigned int hwc_index[MAX_HWCOUNTERS];
	int n_lim;
	int idx;

	if (!ppmu)
		return;
	local_irq_save(flags);
	cpuhw = &__get_cpu_var(cpu_hw_counters);
	if (!cpuhw->disabled) {
		local_irq_restore(flags);
		return;
	}
	cpuhw->disabled = 0;

	/*
	 * If we didn't change anything, or only removed counters,
	 * no need to recalculate MMCR* settings and reset the PMCs.
	 * Just reenable the PMU with the current MMCR* settings
	 * (possibly updated for removal of counters).
	 */
	if (!cpuhw->n_added) {
		mtspr(SPRN_MMCRA, cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE);
		mtspr(SPRN_MMCR1, cpuhw->mmcr[1]);
		if (cpuhw->n_counters == 0)
			ppc_set_pmu_inuse(0);
		goto out_enable;
	}

	/*
	 * Compute MMCR* values for the new set of counters
	 */
	if (ppmu->compute_mmcr(cpuhw->events, cpuhw->n_counters, hwc_index,
			       cpuhw->mmcr)) {
		/* shouldn't ever get here */
		printk(KERN_ERR "oops compute_mmcr failed\n");
		goto out;
	}

	/*
	 * Add in MMCR0 freeze bits corresponding to the
	 * attr.exclude_* bits for the first counter.
	 * We have already checked that all counters have the
	 * same values for these bits as the first counter.
	 */
	counter = cpuhw->counter[0];
	if (counter->attr.exclude_user)
		cpuhw->mmcr[0] |= MMCR0_FCP;
	if (counter->attr.exclude_kernel)
		cpuhw->mmcr[0] |= freeze_counters_kernel;
	if (counter->attr.exclude_hv)
		cpuhw->mmcr[0] |= MMCR0_FCHV;

	/*
	 * Write the new configuration to MMCR* with the freeze
	 * bit set and set the hardware counters to their initial values.
	 * Then unfreeze the counters.
	 */
	ppc_set_pmu_inuse(1);
	mtspr(SPRN_MMCRA, cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE);
	mtspr(SPRN_MMCR1, cpuhw->mmcr[1]);
	mtspr(SPRN_MMCR0, (cpuhw->mmcr[0] & ~(MMCR0_PMC1CE | MMCR0_PMCjCE))
				| MMCR0_FC);

	/*
	 * Read off any pre-existing counters that need to move
	 * to another PMC.
	 */
	for (i = 0; i < cpuhw->n_counters; ++i) {
		counter = cpuhw->counter[i];
		if (counter->hw.idx && counter->hw.idx != hwc_index[i] + 1) {
			power_pmu_read(counter);
			write_pmc(counter->hw.idx, 0);
			counter->hw.idx = 0;
		}
	}

	/*
	 * Initialize the PMCs for all the new and moved counters.
	 */
	cpuhw->n_limited = n_lim = 0;
	for (i = 0; i < cpuhw->n_counters; ++i) {
		counter = cpuhw->counter[i];
		if (counter->hw.idx)
			continue;
		idx = hwc_index[i] + 1;
		if (is_limited_pmc(idx)) {
			cpuhw->limited_counter[n_lim] = counter;
			cpuhw->limited_hwidx[n_lim] = idx;
			++n_lim;
			continue;
		}
		val = 0;
		if (counter->hw.sample_period) {
			left = atomic64_read(&counter->hw.period_left);
			if (left < 0x80000000L)
				val = 0x80000000L - left;
		}
		atomic64_set(&counter->hw.prev_count, val);
		counter->hw.idx = idx;
		write_pmc(idx, val);
		perf_counter_update_userpage(counter);
	}
	cpuhw->n_limited = n_lim;
	cpuhw->mmcr[0] |= MMCR0_PMXE | MMCR0_FCECE;

 out_enable:
	mb();
	write_mmcr0(cpuhw, cpuhw->mmcr[0]);

	/*
	 * Enable instruction sampling if necessary
	 */
	if (cpuhw->mmcr[2] & MMCRA_SAMPLE_ENABLE) {
		mb();
		mtspr(SPRN_MMCRA, cpuhw->mmcr[2]);
	}

 out:
	local_irq_restore(flags);
}

static int collect_events(struct perf_counter *group, int max_count,
			  struct perf_counter *ctrs[], u64 *events,
			  unsigned int *flags)
{
	int n = 0;
	struct perf_counter *counter;

	if (!is_software_counter(group)) {
		if (n >= max_count)
			return -1;
		ctrs[n] = group;
		flags[n] = group->hw.counter_base;
		events[n++] = group->hw.config;
	}
	list_for_each_entry(counter, &group->sibling_list, list_entry) {
		if (!is_software_counter(counter) &&
		    counter->state != PERF_COUNTER_STATE_OFF) {
			if (n >= max_count)
				return -1;
			ctrs[n] = counter;
			flags[n] = counter->hw.counter_base;
			events[n++] = counter->hw.config;
		}
	}
	return n;
}

static void counter_sched_in(struct perf_counter *counter, int cpu)
{
	counter->state = PERF_COUNTER_STATE_ACTIVE;
	counter->oncpu = cpu;
	counter->tstamp_running += counter->ctx->time - counter->tstamp_stopped;
	if (is_software_counter(counter))
		counter->pmu->enable(counter);
}

/*
 * Called to enable a whole group of counters.
 * Returns 1 if the group was enabled, or -EAGAIN if it could not be.
 * Assumes the caller has disabled interrupts and has
 * frozen the PMU with hw_perf_save_disable.
 */
int hw_perf_group_sched_in(struct perf_counter *group_leader,
	       struct perf_cpu_context *cpuctx,
	       struct perf_counter_context *ctx, int cpu)
{
	struct cpu_hw_counters *cpuhw;
	long i, n, n0;
	struct perf_counter *sub;

	if (!ppmu)
		return 0;
	cpuhw = &__get_cpu_var(cpu_hw_counters);
	n0 = cpuhw->n_counters;
	n = collect_events(group_leader, ppmu->n_counter - n0,
			   &cpuhw->counter[n0], &cpuhw->events[n0],
			   &cpuhw->flags[n0]);
	if (n < 0)
		return -EAGAIN;
	if (check_excludes(cpuhw->counter, cpuhw->flags, n0, n))
		return -EAGAIN;
	i = power_check_constraints(cpuhw, cpuhw->events, cpuhw->flags, n + n0);
	if (i < 0)
		return -EAGAIN;
	cpuhw->n_counters = n0 + n;
	cpuhw->n_added += n;

	/*
	 * OK, this group can go on; update counter states etc.,
	 * and enable any software counters
	 */
	for (i = n0; i < n0 + n; ++i)
		cpuhw->counter[i]->hw.config = cpuhw->events[i];
	cpuctx->active_oncpu += n;
	n = 1;
	counter_sched_in(group_leader, cpu);
	list_for_each_entry(sub, &group_leader->sibling_list, list_entry) {
		if (sub->state != PERF_COUNTER_STATE_OFF) {
			counter_sched_in(sub, cpu);
			++n;
		}
	}
	ctx->nr_active += n;

	return 1;
}

/*
 * Add a counter to the PMU.
 * If all counters are not already frozen, then we disable and
 * re-enable the PMU in order to get hw_perf_enable to do the
 * actual work of reconfiguring the PMU.
 */
static int power_pmu_enable(struct perf_counter *counter)
{
	struct cpu_hw_counters *cpuhw;
	unsigned long flags;
	int n0;
	int ret = -EAGAIN;

	local_irq_save(flags);
	perf_disable();

	/*
	 * Add the counter to the list (if there is room)
	 * and check whether the total set is still feasible.
	 */
	cpuhw = &__get_cpu_var(cpu_hw_counters);
	n0 = cpuhw->n_counters;
	if (n0 >= ppmu->n_counter)
		goto out;
	cpuhw->counter[n0] = counter;
	cpuhw->events[n0] = counter->hw.config;
	cpuhw->flags[n0] = counter->hw.counter_base;
	if (check_excludes(cpuhw->counter, cpuhw->flags, n0, 1))
		goto out;
	if (power_check_constraints(cpuhw, cpuhw->events, cpuhw->flags, n0 + 1))
		goto out;

	counter->hw.config = cpuhw->events[n0];
	++cpuhw->n_counters;
	++cpuhw->n_added;

	ret = 0;
 out:
	perf_enable();
	local_irq_restore(flags);
	return ret;
}

/*
 * Remove a counter from the PMU.
 */
static void power_pmu_disable(struct perf_counter *counter)
{
	struct cpu_hw_counters *cpuhw;
	long i;
	unsigned long flags;

	local_irq_save(flags);
	perf_disable();

	power_pmu_read(counter);

	cpuhw = &__get_cpu_var(cpu_hw_counters);
	for (i = 0; i < cpuhw->n_counters; ++i) {
		if (counter == cpuhw->counter[i]) {
			while (++i < cpuhw->n_counters)
				cpuhw->counter[i-1] = cpuhw->counter[i];
			--cpuhw->n_counters;
			ppmu->disable_pmc(counter->hw.idx - 1, cpuhw->mmcr);
			if (counter->hw.idx) {
				write_pmc(counter->hw.idx, 0);
				counter->hw.idx = 0;
			}
			perf_counter_update_userpage(counter);
			break;
		}
	}
	for (i = 0; i < cpuhw->n_limited; ++i)
		if (counter == cpuhw->limited_counter[i])
			break;
	if (i < cpuhw->n_limited) {
		while (++i < cpuhw->n_limited) {
			cpuhw->limited_counter[i-1] = cpuhw->limited_counter[i];
			cpuhw->limited_hwidx[i-1] = cpuhw->limited_hwidx[i];
		}
		--cpuhw->n_limited;
	}
	if (cpuhw->n_counters == 0) {
		/* disable exceptions if no counters are running */
		cpuhw->mmcr[0] &= ~(MMCR0_PMXE | MMCR0_FCECE);
	}

	perf_enable();
	local_irq_restore(flags);
}

/*
 * Re-enable interrupts on a counter after they were throttled
 * because they were coming too fast.
 */
static void power_pmu_unthrottle(struct perf_counter *counter)
{
	s64 val, left;
	unsigned long flags;

	if (!counter->hw.idx || !counter->hw.sample_period)
		return;
	local_irq_save(flags);
	perf_disable();
	power_pmu_read(counter);
	left = counter->hw.sample_period;
	counter->hw.last_period = left;
	val = 0;
	if (left < 0x80000000L)
		val = 0x80000000L - left;
	write_pmc(counter->hw.idx, val);
	atomic64_set(&counter->hw.prev_count, val);
	atomic64_set(&counter->hw.period_left, left);
	perf_counter_update_userpage(counter);
	perf_enable();
	local_irq_restore(flags);
}

struct pmu power_pmu = {
	.enable		= power_pmu_enable,
	.disable	= power_pmu_disable,
	.read		= power_pmu_read,
	.unthrottle	= power_pmu_unthrottle,
};

/*
 * Return 1 if we might be able to put counter on a limited PMC,
 * or 0 if not.
 * A counter can only go on a limited PMC if it counts something
 * that a limited PMC can count, doesn't require interrupts, and
 * doesn't exclude any processor mode.
 */
static int can_go_on_limited_pmc(struct perf_counter *counter, u64 ev,
				 unsigned int flags)
{
	int n;
	u64 alt[MAX_EVENT_ALTERNATIVES];

	if (counter->attr.exclude_user
	    || counter->attr.exclude_kernel
	    || counter->attr.exclude_hv
	    || counter->attr.sample_period)
		return 0;

	if (ppmu->limited_pmc_event(ev))
		return 1;

	/*
	 * The requested event isn't on a limited PMC already;
	 * see if any alternative code goes on a limited PMC.
	 */
	if (!ppmu->get_alternatives)
		return 0;

	flags |= PPMU_LIMITED_PMC_OK | PPMU_LIMITED_PMC_REQD;
	n = ppmu->get_alternatives(ev, flags, alt);

	return n > 0;
}

/*
 * Find an alternative event that goes on a normal PMC, if possible,
 * and return the event code, or 0 if there is no such alternative.
 * (Note: event code 0 is "don't count" on all machines.)
 */
static u64 normal_pmc_alternative(u64 ev, unsigned long flags)
{
	u64 alt[MAX_EVENT_ALTERNATIVES];
	int n;

	flags &= ~(PPMU_LIMITED_PMC_OK | PPMU_LIMITED_PMC_REQD);
	n = ppmu->get_alternatives(ev, flags, alt);
	if (!n)
		return 0;
	return alt[0];
}

/* Number of perf_counters counting hardware events */
static atomic_t num_counters;
/* Used to avoid races in calling reserve/release_pmc_hardware */
static DEFINE_MUTEX(pmc_reserve_mutex);

/*
 * Release the PMU if this is the last perf_counter.
 */
static void hw_perf_counter_destroy(struct perf_counter *counter)
{
	if (!atomic_add_unless(&num_counters, -1, 1)) {
		mutex_lock(&pmc_reserve_mutex);
		if (atomic_dec_return(&num_counters) == 0)
			release_pmc_hardware();
		mutex_unlock(&pmc_reserve_mutex);
	}
}

/*
 * Translate a generic cache event config to a raw event code.
 */
static int hw_perf_cache_event(u64 config, u64 *eventp)
{
	unsigned long type, op, result;
	int ev;

	if (!ppmu->cache_events)
		return -EINVAL;

	/* unpack config */
	type = config & 0xff;
	op = (config >> 8) & 0xff;
	result = (config >> 16) & 0xff;

	if (type >= PERF_COUNT_HW_CACHE_MAX ||
	    op >= PERF_COUNT_HW_CACHE_OP_MAX ||
	    result >= PERF_COUNT_HW_CACHE_RESULT_MAX)
		return -EINVAL;

	ev = (*ppmu->cache_events)[type][op][result];
	if (ev == 0)
		return -EOPNOTSUPP;
	if (ev == -1)
		return -EINVAL;
	*eventp = ev;
	return 0;
}

const struct pmu *hw_perf_counter_init(struct perf_counter *counter)
{
	u64 ev;
	unsigned long flags;
	struct perf_counter *ctrs[MAX_HWCOUNTERS];
	u64 events[MAX_HWCOUNTERS];
	unsigned int cflags[MAX_HWCOUNTERS];
	int n;
	int err;
	struct cpu_hw_counters *cpuhw;

	if (!ppmu)
		return ERR_PTR(-ENXIO);
	switch (counter->attr.type) {
	case PERF_TYPE_HARDWARE:
		ev = counter->attr.config;
		if (ev >= ppmu->n_generic || ppmu->generic_events[ev] == 0)
			return ERR_PTR(-EOPNOTSUPP);
		ev = ppmu->generic_events[ev];
		break;
	case PERF_TYPE_HW_CACHE:
		err = hw_perf_cache_event(counter->attr.config, &ev);
		if (err)
			return ERR_PTR(err);
		break;
	case PERF_TYPE_RAW:
		ev = counter->attr.config;
		break;
	default:
		return ERR_PTR(-EINVAL);
	}
	counter->hw.config_base = ev;
	counter->hw.idx = 0;

	/*
	 * If we are not running on a hypervisor, force the
	 * exclude_hv bit to 0 so that we don't care what
	 * the user set it to.
	 */
	if (!firmware_has_feature(FW_FEATURE_LPAR))
		counter->attr.exclude_hv = 0;

	/*
	 * If this is a per-task counter, then we can use
	 * PM_RUN_* events interchangeably with their non RUN_*
	 * equivalents, e.g. PM_RUN_CYC instead of PM_CYC.
	 * XXX we should check if the task is an idle task.
	 */
	flags = 0;
	if (counter->ctx->task)
		flags |= PPMU_ONLY_COUNT_RUN;

	/*
	 * If this machine has limited counters, check whether this
	 * event could go on a limited counter.
	 */
	if (ppmu->flags & PPMU_LIMITED_PMC5_6) {
		if (can_go_on_limited_pmc(counter, ev, flags)) {
			flags |= PPMU_LIMITED_PMC_OK;
		} else if (ppmu->limited_pmc_event(ev)) {
			/*
			 * The requested event is on a limited PMC,
			 * but we can't use a limited PMC; see if any
			 * alternative goes on a normal PMC.
			 */
			ev = normal_pmc_alternative(ev, flags);
			if (!ev)
				return ERR_PTR(-EINVAL);
		}
	}

	/*
	 * If this is in a group, check if it can go on with all the
	 * other hardware counters in the group.  We assume the counter
	 * hasn't been linked into its leader's sibling list at this point.
	 */
	n = 0;
	if (counter->group_leader != counter) {
		n = collect_events(counter->group_leader, ppmu->n_counter - 1,
				   ctrs, events, cflags);
		if (n < 0)
			return ERR_PTR(-EINVAL);
	}
	events[n] = ev;
	ctrs[n] = counter;
	cflags[n] = flags;
	if (check_excludes(ctrs, cflags, n, 1))
		return ERR_PTR(-EINVAL);

	cpuhw = &get_cpu_var(cpu_hw_counters);
	err = power_check_constraints(cpuhw, events, cflags, n + 1);
	put_cpu_var(cpu_hw_counters);
	if (err)
		return ERR_PTR(-EINVAL);

	counter->hw.config = events[n];
	counter->hw.counter_base = cflags[n];
	counter->hw.last_period = counter->hw.sample_period;
	atomic64_set(&counter->hw.period_left, counter->hw.last_period);

	/*
	 * See if we need to reserve the PMU.
	 * If no counters are currently in use, then we have to take a
	 * mutex to ensure that we don't race with another task doing
	 * reserve_pmc_hardware or release_pmc_hardware.
	 */
	err = 0;
	if (!atomic_inc_not_zero(&num_counters)) {
		mutex_lock(&pmc_reserve_mutex);
		if (atomic_read(&num_counters) == 0 &&
		    reserve_pmc_hardware(perf_counter_interrupt))
			err = -EBUSY;
		else
			atomic_inc(&num_counters);
		mutex_unlock(&pmc_reserve_mutex);
	}
	counter->destroy = hw_perf_counter_destroy;

	if (err)
		return ERR_PTR(err);
	return &power_pmu;
}

/*
 * A counter has overflowed; update its count and record
 * things if requested.  Note that interrupts are hard-disabled
 * here so there is no possibility of being interrupted.
 */
static void record_and_restart(struct perf_counter *counter, unsigned long val,
			       struct pt_regs *regs, int nmi)
{
	u64 period = counter->hw.sample_period;
	s64 prev, delta, left;
	int record = 0;

	/* we don't have to worry about interrupts here */
	prev = atomic64_read(&counter->hw.prev_count);
	delta = (val - prev) & 0xfffffffful;
	atomic64_add(delta, &counter->count);

	/*
	 * See if the total period for this counter has expired,
	 * and update for the next period.
	 */
	val = 0;
	left = atomic64_read(&counter->hw.period_left) - delta;
	if (period) {
		if (left <= 0) {
			left += period;
			if (left <= 0)
				left = period;
			record = 1;
		}
		if (left < 0x80000000LL)
			val = 0x80000000LL - left;
	}

	/*
	 * Finally record data if requested.
	 */
	if (record) {
		struct perf_sample_data data = {
			.regs	= regs,
			.addr	= 0,
			.period	= counter->hw.last_period,
		};

		if (counter->attr.sample_type & PERF_SAMPLE_ADDR)
			perf_get_data_addr(regs, &data.addr);

		if (perf_counter_overflow(counter, nmi, &data)) {
			/*
			 * Interrupts are coming too fast - throttle them
			 * by setting the counter to 0, so it will be
			 * at least 2^30 cycles until the next interrupt
			 * (assuming each counter counts at most 2 counts
			 * per cycle).
			 */
			val = 0;
			left = ~0ULL >> 1;
		}
	}

	write_pmc(counter->hw.idx, val);
	atomic64_set(&counter->hw.prev_count, val);
	atomic64_set(&counter->hw.period_left, left);
	perf_counter_update_userpage(counter);
}

/*
 * Called from generic code to get the misc flags (i.e. processor mode)
 * for an event.
 */
unsigned long perf_misc_flags(struct pt_regs *regs)
{
	u32 flags = perf_get_misc_flags(regs);

	if (flags)
		return flags;
	return user_mode(regs) ? PERF_EVENT_MISC_USER :
		PERF_EVENT_MISC_KERNEL;
}

/*
 * Called from generic code to get the instruction pointer
 * for an event.
 */
unsigned long perf_instruction_pointer(struct pt_regs *regs)
{
	unsigned long ip;

	if (TRAP(regs) != 0xf00)
		return regs->nip;	/* not a PMU interrupt */

	ip = mfspr(SPRN_SIAR) + perf_ip_adjust(regs);
	return ip;
}

/*
 * Performance monitor interrupt stuff
 */
static void perf_counter_interrupt(struct pt_regs *regs)
{
	int i;
	struct cpu_hw_counters *cpuhw = &__get_cpu_var(cpu_hw_counters);
	struct perf_counter *counter;
	unsigned long val;
	int found = 0;
	int nmi;

	if (cpuhw->n_limited)
		freeze_limited_counters(cpuhw, mfspr(SPRN_PMC5),
					mfspr(SPRN_PMC6));

	perf_read_regs(regs);

	nmi = perf_intr_is_nmi(regs);
	if (nmi)
		nmi_enter();
	else
		irq_enter();

	for (i = 0; i < cpuhw->n_counters; ++i) {
		counter = cpuhw->counter[i];
		if (!counter->hw.idx || is_limited_pmc(counter->hw.idx))
			continue;
		val = read_pmc(counter->hw.idx);
		if ((int)val < 0) {
			/* counter has overflowed */
			found = 1;
			record_and_restart(counter, val, regs, nmi);
		}
	}

	/*
	 * In case we didn't find and reset the counter that caused
	 * the interrupt, scan all counters and reset any that are
	 * negative, to avoid getting continual interrupts.
	 * Any that we processed in the previous loop will not be negative.
	 */
	if (!found) {
		for (i = 0; i < ppmu->n_counter; ++i) {
			if (is_limited_pmc(i + 1))
				continue;
			val = read_pmc(i + 1);
			if ((int)val < 0)
				write_pmc(i + 1, 0);
		}
	}

	/*
	 * Reset MMCR0 to its normal value.  This will set PMXE and
	 * clear FC (freeze counters) and PMAO (perf mon alert occurred)
	 * and thus allow interrupts to occur again.
	 * XXX might want to use MSR.PM to keep the counters frozen until
	 * we get back out of this interrupt.
	 */
	write_mmcr0(cpuhw, cpuhw->mmcr[0]);

	if (nmi)
		nmi_exit();
	else
		irq_exit();
}

void hw_perf_counter_setup(int cpu)
{
	struct cpu_hw_counters *cpuhw = &per_cpu(cpu_hw_counters, cpu);

	if (!ppmu)
		return;
	memset(cpuhw, 0, sizeof(*cpuhw));
	cpuhw->mmcr[0] = MMCR0_FC;
}

int register_power_pmu(struct power_pmu *pmu)
{
	if (ppmu)
		return -EBUSY;		/* something's already registered */

	ppmu = pmu;
	pr_info("%s performance monitor hardware support registered\n",
		pmu->name);

#ifdef MSR_HV
	/*
	 * Use FCHV to ignore kernel events if MSR.HV is set.
	 */
	if (mfmsr() & MSR_HV)
		freeze_counters_kernel = MMCR0_FCHV;
#endif /* CONFIG_PPC64 */

	return 0;
}