aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86')
-rw-r--r--arch/x86/include/asm/perf_event_p4.h52
-rw-r--r--arch/x86/kernel/cpu/perf_event.c259
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel.c8
-rw-r--r--arch/x86/kernel/cpu/perf_event_intel_ds.c13
-rw-r--r--arch/x86/kernel/cpu/perf_event_p4.c292
5 files changed, 424 insertions, 200 deletions
diff --git a/arch/x86/include/asm/perf_event_p4.h b/arch/x86/include/asm/perf_event_p4.h
index def500776b16..a70cd216be5d 100644
--- a/arch/x86/include/asm/perf_event_p4.h
+++ b/arch/x86/include/asm/perf_event_p4.h
@@ -36,19 +36,6 @@
36#define P4_ESCR_EMASK(v) ((v) << P4_ESCR_EVENTMASK_SHIFT) 36#define P4_ESCR_EMASK(v) ((v) << P4_ESCR_EVENTMASK_SHIFT)
37#define P4_ESCR_TAG(v) ((v) << P4_ESCR_TAG_SHIFT) 37#define P4_ESCR_TAG(v) ((v) << P4_ESCR_TAG_SHIFT)
38 38
39/* Non HT mask */
40#define P4_ESCR_MASK \
41 (P4_ESCR_EVENT_MASK | \
42 P4_ESCR_EVENTMASK_MASK | \
43 P4_ESCR_TAG_MASK | \
44 P4_ESCR_TAG_ENABLE | \
45 P4_ESCR_T0_OS | \
46 P4_ESCR_T0_USR)
47
48/* HT mask */
49#define P4_ESCR_MASK_HT \
50 (P4_ESCR_MASK | P4_ESCR_T1_OS | P4_ESCR_T1_USR)
51
52#define P4_CCCR_OVF 0x80000000U 39#define P4_CCCR_OVF 0x80000000U
53#define P4_CCCR_CASCADE 0x40000000U 40#define P4_CCCR_CASCADE 0x40000000U
54#define P4_CCCR_OVF_PMI_T0 0x04000000U 41#define P4_CCCR_OVF_PMI_T0 0x04000000U
@@ -70,23 +57,6 @@
70#define P4_CCCR_THRESHOLD(v) ((v) << P4_CCCR_THRESHOLD_SHIFT) 57#define P4_CCCR_THRESHOLD(v) ((v) << P4_CCCR_THRESHOLD_SHIFT)
71#define P4_CCCR_ESEL(v) ((v) << P4_CCCR_ESCR_SELECT_SHIFT) 58#define P4_CCCR_ESEL(v) ((v) << P4_CCCR_ESCR_SELECT_SHIFT)
72 59
73/* Non HT mask */
74#define P4_CCCR_MASK \
75 (P4_CCCR_OVF | \
76 P4_CCCR_CASCADE | \
77 P4_CCCR_OVF_PMI_T0 | \
78 P4_CCCR_FORCE_OVF | \
79 P4_CCCR_EDGE | \
80 P4_CCCR_THRESHOLD_MASK | \
81 P4_CCCR_COMPLEMENT | \
82 P4_CCCR_COMPARE | \
83 P4_CCCR_ESCR_SELECT_MASK | \
84 P4_CCCR_ENABLE)
85
86/* HT mask */
87#define P4_CCCR_MASK_HT \
88 (P4_CCCR_MASK | P4_CCCR_OVF_PMI_T1 | P4_CCCR_THREAD_ANY)
89
90#define P4_GEN_ESCR_EMASK(class, name, bit) \ 60#define P4_GEN_ESCR_EMASK(class, name, bit) \
91 class##__##name = ((1 << bit) << P4_ESCR_EVENTMASK_SHIFT) 61 class##__##name = ((1 << bit) << P4_ESCR_EVENTMASK_SHIFT)
92#define P4_ESCR_EMASK_BIT(class, name) class##__##name 62#define P4_ESCR_EMASK_BIT(class, name) class##__##name
@@ -127,6 +97,28 @@
127#define P4_CONFIG_HT_SHIFT 63 97#define P4_CONFIG_HT_SHIFT 63
128#define P4_CONFIG_HT (1ULL << P4_CONFIG_HT_SHIFT) 98#define P4_CONFIG_HT (1ULL << P4_CONFIG_HT_SHIFT)
129 99
100/*
101 * The bits we allow to pass for RAW events
102 */
103#define P4_CONFIG_MASK_ESCR \
104 P4_ESCR_EVENT_MASK | \
105 P4_ESCR_EVENTMASK_MASK | \
106 P4_ESCR_TAG_MASK | \
107 P4_ESCR_TAG_ENABLE
108
109#define P4_CONFIG_MASK_CCCR \
110 P4_CCCR_EDGE | \
111 P4_CCCR_THRESHOLD_MASK | \
112 P4_CCCR_COMPLEMENT | \
113 P4_CCCR_COMPARE | \
114 P4_CCCR_THREAD_ANY | \
115 P4_CCCR_RESERVED
116
117/* some dangerous bits are reserved for kernel internals */
118#define P4_CONFIG_MASK \
119 (p4_config_pack_escr(P4_CONFIG_MASK_ESCR)) | \
120 (p4_config_pack_cccr(P4_CONFIG_MASK_CCCR))
121
130static inline bool p4_is_event_cascaded(u64 config) 122static inline bool p4_is_event_cascaded(u64 config)
131{ 123{
132 u32 cccr = p4_config_unpack_cccr(config); 124 u32 cccr = p4_config_unpack_cccr(config);
diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c
index 3efdf2870a35..0fb17050360f 100644
--- a/arch/x86/kernel/cpu/perf_event.c
+++ b/arch/x86/kernel/cpu/perf_event.c
@@ -530,7 +530,7 @@ static int x86_pmu_hw_config(struct perf_event *event)
530/* 530/*
531 * Setup the hardware configuration for a given attr_type 531 * Setup the hardware configuration for a given attr_type
532 */ 532 */
533static int __hw_perf_event_init(struct perf_event *event) 533static int __x86_pmu_event_init(struct perf_event *event)
534{ 534{
535 int err; 535 int err;
536 536
@@ -583,7 +583,7 @@ static void x86_pmu_disable_all(void)
583 } 583 }
584} 584}
585 585
586void hw_perf_disable(void) 586static void x86_pmu_disable(struct pmu *pmu)
587{ 587{
588 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); 588 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
589 589
@@ -618,7 +618,7 @@ static void x86_pmu_enable_all(int added)
618 } 618 }
619} 619}
620 620
621static const struct pmu pmu; 621static struct pmu pmu;
622 622
623static inline int is_x86_event(struct perf_event *event) 623static inline int is_x86_event(struct perf_event *event)
624{ 624{
@@ -800,10 +800,10 @@ static inline int match_prev_assignment(struct hw_perf_event *hwc,
800 hwc->last_tag == cpuc->tags[i]; 800 hwc->last_tag == cpuc->tags[i];
801} 801}
802 802
803static int x86_pmu_start(struct perf_event *event); 803static void x86_pmu_start(struct perf_event *event, int flags);
804static void x86_pmu_stop(struct perf_event *event); 804static void x86_pmu_stop(struct perf_event *event, int flags);
805 805
806void hw_perf_enable(void) 806static void x86_pmu_enable(struct pmu *pmu)
807{ 807{
808 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); 808 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
809 struct perf_event *event; 809 struct perf_event *event;
@@ -839,7 +839,14 @@ void hw_perf_enable(void)
839 match_prev_assignment(hwc, cpuc, i)) 839 match_prev_assignment(hwc, cpuc, i))
840 continue; 840 continue;
841 841
842 x86_pmu_stop(event); 842 /*
843 * Ensure we don't accidentally enable a stopped
844 * counter simply because we rescheduled.
845 */
846 if (hwc->state & PERF_HES_STOPPED)
847 hwc->state |= PERF_HES_ARCH;
848
849 x86_pmu_stop(event, PERF_EF_UPDATE);
843 } 850 }
844 851
845 for (i = 0; i < cpuc->n_events; i++) { 852 for (i = 0; i < cpuc->n_events; i++) {
@@ -851,7 +858,10 @@ void hw_perf_enable(void)
851 else if (i < n_running) 858 else if (i < n_running)
852 continue; 859 continue;
853 860
854 x86_pmu_start(event); 861 if (hwc->state & PERF_HES_ARCH)
862 continue;
863
864 x86_pmu_start(event, PERF_EF_RELOAD);
855 } 865 }
856 cpuc->n_added = 0; 866 cpuc->n_added = 0;
857 perf_events_lapic_init(); 867 perf_events_lapic_init();
@@ -952,15 +962,12 @@ static void x86_pmu_enable_event(struct perf_event *event)
952} 962}
953 963
954/* 964/*
955 * activate a single event 965 * Add a single event to the PMU.
956 * 966 *
957 * The event is added to the group of enabled events 967 * The event is added to the group of enabled events
958 * but only if it can be scehduled with existing events. 968 * but only if it can be scehduled with existing events.
959 *
960 * Called with PMU disabled. If successful and return value 1,
961 * then guaranteed to call perf_enable() and hw_perf_enable()
962 */ 969 */
963static int x86_pmu_enable(struct perf_event *event) 970static int x86_pmu_add(struct perf_event *event, int flags)
964{ 971{
965 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); 972 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
966 struct hw_perf_event *hwc; 973 struct hw_perf_event *hwc;
@@ -969,57 +976,66 @@ static int x86_pmu_enable(struct perf_event *event)
969 976
970 hwc = &event->hw; 977 hwc = &event->hw;
971 978
979 perf_pmu_disable(event->pmu);
972 n0 = cpuc->n_events; 980 n0 = cpuc->n_events;
973 n = collect_events(cpuc, event, false); 981 ret = n = collect_events(cpuc, event, false);
974 if (n < 0) 982 if (ret < 0)
975 return n; 983 goto out;
984
985 hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
986 if (!(flags & PERF_EF_START))
987 hwc->state |= PERF_HES_ARCH;
976 988
977 /* 989 /*
978 * If group events scheduling transaction was started, 990 * If group events scheduling transaction was started,
979 * skip the schedulability test here, it will be peformed 991 * skip the schedulability test here, it will be peformed
980 * at commit time(->commit_txn) as a whole 992 * at commit time (->commit_txn) as a whole
981 */ 993 */
982 if (cpuc->group_flag & PERF_EVENT_TXN) 994 if (cpuc->group_flag & PERF_EVENT_TXN)
983 goto out; 995 goto done_collect;
984 996
985 ret = x86_pmu.schedule_events(cpuc, n, assign); 997 ret = x86_pmu.schedule_events(cpuc, n, assign);
986 if (ret) 998 if (ret)
987 return ret; 999 goto out;
988 /* 1000 /*
989 * copy new assignment, now we know it is possible 1001 * copy new assignment, now we know it is possible
990 * will be used by hw_perf_enable() 1002 * will be used by hw_perf_enable()
991 */ 1003 */
992 memcpy(cpuc->assign, assign, n*sizeof(int)); 1004 memcpy(cpuc->assign, assign, n*sizeof(int));
993 1005
994out: 1006done_collect:
995 cpuc->n_events = n; 1007 cpuc->n_events = n;
996 cpuc->n_added += n - n0; 1008 cpuc->n_added += n - n0;
997 cpuc->n_txn += n - n0; 1009 cpuc->n_txn += n - n0;
998 1010
999 return 0; 1011 ret = 0;
1012out:
1013 perf_pmu_enable(event->pmu);
1014 return ret;
1000} 1015}
1001 1016
1002static int x86_pmu_start(struct perf_event *event) 1017static void x86_pmu_start(struct perf_event *event, int flags)
1003{ 1018{
1004 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); 1019 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
1005 int idx = event->hw.idx; 1020 int idx = event->hw.idx;
1006 1021
1007 if (idx == -1) 1022 if (WARN_ON_ONCE(!(event->hw.state & PERF_HES_STOPPED)))
1008 return -EAGAIN; 1023 return;
1024
1025 if (WARN_ON_ONCE(idx == -1))
1026 return;
1027
1028 if (flags & PERF_EF_RELOAD) {
1029 WARN_ON_ONCE(!(event->hw.state & PERF_HES_UPTODATE));
1030 x86_perf_event_set_period(event);
1031 }
1032
1033 event->hw.state = 0;
1009 1034
1010 x86_perf_event_set_period(event);
1011 cpuc->events[idx] = event; 1035 cpuc->events[idx] = event;
1012 __set_bit(idx, cpuc->active_mask); 1036 __set_bit(idx, cpuc->active_mask);
1013 x86_pmu.enable(event); 1037 x86_pmu.enable(event);
1014 perf_event_update_userpage(event); 1038 perf_event_update_userpage(event);
1015
1016 return 0;
1017}
1018
1019static void x86_pmu_unthrottle(struct perf_event *event)
1020{
1021 int ret = x86_pmu_start(event);
1022 WARN_ON_ONCE(ret);
1023} 1039}
1024 1040
1025void perf_event_print_debug(void) 1041void perf_event_print_debug(void)
@@ -1076,27 +1092,29 @@ void perf_event_print_debug(void)
1076 local_irq_restore(flags); 1092 local_irq_restore(flags);
1077} 1093}
1078 1094
1079static void x86_pmu_stop(struct perf_event *event) 1095static void x86_pmu_stop(struct perf_event *event, int flags)
1080{ 1096{
1081 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); 1097 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
1082 struct hw_perf_event *hwc = &event->hw; 1098 struct hw_perf_event *hwc = &event->hw;
1083 int idx = hwc->idx;
1084 1099
1085 if (!__test_and_clear_bit(idx, cpuc->active_mask)) 1100 if (__test_and_clear_bit(hwc->idx, cpuc->active_mask)) {
1086 return; 1101 x86_pmu.disable(event);
1087 1102 cpuc->events[hwc->idx] = NULL;
1088 x86_pmu.disable(event); 1103 WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
1089 1104 hwc->state |= PERF_HES_STOPPED;
1090 /* 1105 }
1091 * Drain the remaining delta count out of a event
1092 * that we are disabling:
1093 */
1094 x86_perf_event_update(event);
1095 1106
1096 cpuc->events[idx] = NULL; 1107 if ((flags & PERF_EF_UPDATE) && !(hwc->state & PERF_HES_UPTODATE)) {
1108 /*
1109 * Drain the remaining delta count out of a event
1110 * that we are disabling:
1111 */
1112 x86_perf_event_update(event);
1113 hwc->state |= PERF_HES_UPTODATE;
1114 }
1097} 1115}
1098 1116
1099static void x86_pmu_disable(struct perf_event *event) 1117static void x86_pmu_del(struct perf_event *event, int flags)
1100{ 1118{
1101 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); 1119 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
1102 int i; 1120 int i;
@@ -1109,7 +1127,7 @@ static void x86_pmu_disable(struct perf_event *event)
1109 if (cpuc->group_flag & PERF_EVENT_TXN) 1127 if (cpuc->group_flag & PERF_EVENT_TXN)
1110 return; 1128 return;
1111 1129
1112 x86_pmu_stop(event); 1130 x86_pmu_stop(event, PERF_EF_UPDATE);
1113 1131
1114 for (i = 0; i < cpuc->n_events; i++) { 1132 for (i = 0; i < cpuc->n_events; i++) {
1115 if (event == cpuc->event_list[i]) { 1133 if (event == cpuc->event_list[i]) {
@@ -1161,7 +1179,7 @@ static int x86_pmu_handle_irq(struct pt_regs *regs)
1161 continue; 1179 continue;
1162 1180
1163 if (perf_event_overflow(event, 1, &data, regs)) 1181 if (perf_event_overflow(event, 1, &data, regs))
1164 x86_pmu_stop(event); 1182 x86_pmu_stop(event, 0);
1165 } 1183 }
1166 1184
1167 if (handled) 1185 if (handled)
@@ -1378,7 +1396,6 @@ void __init init_hw_perf_events(void)
1378 x86_pmu.num_counters = X86_PMC_MAX_GENERIC; 1396 x86_pmu.num_counters = X86_PMC_MAX_GENERIC;
1379 } 1397 }
1380 x86_pmu.intel_ctrl = (1 << x86_pmu.num_counters) - 1; 1398 x86_pmu.intel_ctrl = (1 << x86_pmu.num_counters) - 1;
1381 perf_max_events = x86_pmu.num_counters;
1382 1399
1383 if (x86_pmu.num_counters_fixed > X86_PMC_MAX_FIXED) { 1400 if (x86_pmu.num_counters_fixed > X86_PMC_MAX_FIXED) {
1384 WARN(1, KERN_ERR "hw perf events fixed %d > max(%d), clipping!", 1401 WARN(1, KERN_ERR "hw perf events fixed %d > max(%d), clipping!",
@@ -1414,6 +1431,7 @@ void __init init_hw_perf_events(void)
1414 pr_info("... fixed-purpose events: %d\n", x86_pmu.num_counters_fixed); 1431 pr_info("... fixed-purpose events: %d\n", x86_pmu.num_counters_fixed);
1415 pr_info("... event mask: %016Lx\n", x86_pmu.intel_ctrl); 1432 pr_info("... event mask: %016Lx\n", x86_pmu.intel_ctrl);
1416 1433
1434 perf_pmu_register(&pmu);
1417 perf_cpu_notifier(x86_pmu_notifier); 1435 perf_cpu_notifier(x86_pmu_notifier);
1418} 1436}
1419 1437
@@ -1427,10 +1445,11 @@ static inline void x86_pmu_read(struct perf_event *event)
1427 * Set the flag to make pmu::enable() not perform the 1445 * Set the flag to make pmu::enable() not perform the
1428 * schedulability test, it will be performed at commit time 1446 * schedulability test, it will be performed at commit time
1429 */ 1447 */
1430static void x86_pmu_start_txn(const struct pmu *pmu) 1448static void x86_pmu_start_txn(struct pmu *pmu)
1431{ 1449{
1432 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); 1450 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
1433 1451
1452 perf_pmu_disable(pmu);
1434 cpuc->group_flag |= PERF_EVENT_TXN; 1453 cpuc->group_flag |= PERF_EVENT_TXN;
1435 cpuc->n_txn = 0; 1454 cpuc->n_txn = 0;
1436} 1455}
@@ -1440,7 +1459,7 @@ static void x86_pmu_start_txn(const struct pmu *pmu)
1440 * Clear the flag and pmu::enable() will perform the 1459 * Clear the flag and pmu::enable() will perform the
1441 * schedulability test. 1460 * schedulability test.
1442 */ 1461 */
1443static void x86_pmu_cancel_txn(const struct pmu *pmu) 1462static void x86_pmu_cancel_txn(struct pmu *pmu)
1444{ 1463{
1445 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); 1464 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
1446 1465
@@ -1450,6 +1469,7 @@ static void x86_pmu_cancel_txn(const struct pmu *pmu)
1450 */ 1469 */
1451 cpuc->n_added -= cpuc->n_txn; 1470 cpuc->n_added -= cpuc->n_txn;
1452 cpuc->n_events -= cpuc->n_txn; 1471 cpuc->n_events -= cpuc->n_txn;
1472 perf_pmu_enable(pmu);
1453} 1473}
1454 1474
1455/* 1475/*
@@ -1457,7 +1477,7 @@ static void x86_pmu_cancel_txn(const struct pmu *pmu)
1457 * Perform the group schedulability test as a whole 1477 * Perform the group schedulability test as a whole
1458 * Return 0 if success 1478 * Return 0 if success
1459 */ 1479 */
1460static int x86_pmu_commit_txn(const struct pmu *pmu) 1480static int x86_pmu_commit_txn(struct pmu *pmu)
1461{ 1481{
1462 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); 1482 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
1463 int assign[X86_PMC_IDX_MAX]; 1483 int assign[X86_PMC_IDX_MAX];
@@ -1479,22 +1499,10 @@ static int x86_pmu_commit_txn(const struct pmu *pmu)
1479 memcpy(cpuc->assign, assign, n*sizeof(int)); 1499 memcpy(cpuc->assign, assign, n*sizeof(int));
1480 1500
1481 cpuc->group_flag &= ~PERF_EVENT_TXN; 1501 cpuc->group_flag &= ~PERF_EVENT_TXN;
1482 1502 perf_pmu_enable(pmu);
1483 return 0; 1503 return 0;
1484} 1504}
1485 1505
1486static const struct pmu pmu = {
1487 .enable = x86_pmu_enable,
1488 .disable = x86_pmu_disable,
1489 .start = x86_pmu_start,
1490 .stop = x86_pmu_stop,
1491 .read = x86_pmu_read,
1492 .unthrottle = x86_pmu_unthrottle,
1493 .start_txn = x86_pmu_start_txn,
1494 .cancel_txn = x86_pmu_cancel_txn,
1495 .commit_txn = x86_pmu_commit_txn,
1496};
1497
1498/* 1506/*
1499 * validate that we can schedule this event 1507 * validate that we can schedule this event
1500 */ 1508 */
@@ -1569,12 +1577,22 @@ out:
1569 return ret; 1577 return ret;
1570} 1578}
1571 1579
1572const struct pmu *hw_perf_event_init(struct perf_event *event) 1580int x86_pmu_event_init(struct perf_event *event)
1573{ 1581{
1574 const struct pmu *tmp; 1582 struct pmu *tmp;
1575 int err; 1583 int err;
1576 1584
1577 err = __hw_perf_event_init(event); 1585 switch (event->attr.type) {
1586 case PERF_TYPE_RAW:
1587 case PERF_TYPE_HARDWARE:
1588 case PERF_TYPE_HW_CACHE:
1589 break;
1590
1591 default:
1592 return -ENOENT;
1593 }
1594
1595 err = __x86_pmu_event_init(event);
1578 if (!err) { 1596 if (!err) {
1579 /* 1597 /*
1580 * we temporarily connect event to its pmu 1598 * we temporarily connect event to its pmu
@@ -1594,26 +1612,31 @@ const struct pmu *hw_perf_event_init(struct perf_event *event)
1594 if (err) { 1612 if (err) {
1595 if (event->destroy) 1613 if (event->destroy)
1596 event->destroy(event); 1614 event->destroy(event);
1597 return ERR_PTR(err);
1598 } 1615 }
1599 1616
1600 return &pmu; 1617 return err;
1601} 1618}
1602 1619
1603/* 1620static struct pmu pmu = {
1604 * callchain support 1621 .pmu_enable = x86_pmu_enable,
1605 */ 1622 .pmu_disable = x86_pmu_disable,
1606 1623
1607static inline 1624 .event_init = x86_pmu_event_init,
1608void callchain_store(struct perf_callchain_entry *entry, u64 ip)
1609{
1610 if (entry->nr < PERF_MAX_STACK_DEPTH)
1611 entry->ip[entry->nr++] = ip;
1612}
1613 1625
1614static DEFINE_PER_CPU(struct perf_callchain_entry, pmc_irq_entry); 1626 .add = x86_pmu_add,
1615static DEFINE_PER_CPU(struct perf_callchain_entry, pmc_nmi_entry); 1627 .del = x86_pmu_del,
1628 .start = x86_pmu_start,
1629 .stop = x86_pmu_stop,
1630 .read = x86_pmu_read,
1631
1632 .start_txn = x86_pmu_start_txn,
1633 .cancel_txn = x86_pmu_cancel_txn,
1634 .commit_txn = x86_pmu_commit_txn,
1635};
1616 1636
1637/*
1638 * callchain support
1639 */
1617 1640
1618static void 1641static void
1619backtrace_warning_symbol(void *data, char *msg, unsigned long symbol) 1642backtrace_warning_symbol(void *data, char *msg, unsigned long symbol)
@@ -1635,7 +1658,7 @@ static void backtrace_address(void *data, unsigned long addr, int reliable)
1635{ 1658{
1636 struct perf_callchain_entry *entry = data; 1659 struct perf_callchain_entry *entry = data;
1637 1660
1638 callchain_store(entry, addr); 1661 perf_callchain_store(entry, addr);
1639} 1662}
1640 1663
1641static const struct stacktrace_ops backtrace_ops = { 1664static const struct stacktrace_ops backtrace_ops = {
@@ -1646,11 +1669,15 @@ static const struct stacktrace_ops backtrace_ops = {
1646 .walk_stack = print_context_stack_bp, 1669 .walk_stack = print_context_stack_bp,
1647}; 1670};
1648 1671
1649static void 1672void
1650perf_callchain_kernel(struct pt_regs *regs, struct perf_callchain_entry *entry) 1673perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs)
1651{ 1674{
1652 callchain_store(entry, PERF_CONTEXT_KERNEL); 1675 if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
1653 callchain_store(entry, regs->ip); 1676 /* TODO: We don't support guest os callchain now */
1677 return;
1678 }
1679
1680 perf_callchain_store(entry, regs->ip);
1654 1681
1655 dump_trace(NULL, regs, NULL, regs->bp, &backtrace_ops, entry); 1682 dump_trace(NULL, regs, NULL, regs->bp, &backtrace_ops, entry);
1656} 1683}
@@ -1679,7 +1706,7 @@ perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry *entry)
1679 if (fp < compat_ptr(regs->sp)) 1706 if (fp < compat_ptr(regs->sp))
1680 break; 1707 break;
1681 1708
1682 callchain_store(entry, frame.return_address); 1709 perf_callchain_store(entry, frame.return_address);
1683 fp = compat_ptr(frame.next_frame); 1710 fp = compat_ptr(frame.next_frame);
1684 } 1711 }
1685 return 1; 1712 return 1;
@@ -1692,19 +1719,20 @@ perf_callchain_user32(struct pt_regs *regs, struct perf_callchain_entry *entry)
1692} 1719}
1693#endif 1720#endif
1694 1721
1695static void 1722void
1696perf_callchain_user(struct pt_regs *regs, struct perf_callchain_entry *entry) 1723perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
1697{ 1724{
1698 struct stack_frame frame; 1725 struct stack_frame frame;
1699 const void __user *fp; 1726 const void __user *fp;
1700 1727
1701 if (!user_mode(regs)) 1728 if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
1702 regs = task_pt_regs(current); 1729 /* TODO: We don't support guest os callchain now */
1730 return;
1731 }
1703 1732
1704 fp = (void __user *)regs->bp; 1733 fp = (void __user *)regs->bp;
1705 1734
1706 callchain_store(entry, PERF_CONTEXT_USER); 1735 perf_callchain_store(entry, regs->ip);
1707 callchain_store(entry, regs->ip);
1708 1736
1709 if (perf_callchain_user32(regs, entry)) 1737 if (perf_callchain_user32(regs, entry))
1710 return; 1738 return;
@@ -1721,52 +1749,11 @@ perf_callchain_user(struct pt_regs *regs, struct perf_callchain_entry *entry)
1721 if ((unsigned long)fp < regs->sp) 1749 if ((unsigned long)fp < regs->sp)
1722 break; 1750 break;
1723 1751
1724 callchain_store(entry, frame.return_address); 1752 perf_callchain_store(entry, frame.return_address);
1725 fp = frame.next_frame; 1753 fp = frame.next_frame;
1726 } 1754 }
1727} 1755}
1728 1756
1729static void
1730perf_do_callchain(struct pt_regs *regs, struct perf_callchain_entry *entry)
1731{
1732 int is_user;
1733
1734 if (!regs)
1735 return;
1736
1737 is_user = user_mode(regs);
1738
1739 if (is_user && current->state != TASK_RUNNING)
1740 return;
1741
1742 if (!is_user)
1743 perf_callchain_kernel(regs, entry);
1744
1745 if (current->mm)
1746 perf_callchain_user(regs, entry);
1747}
1748
1749struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
1750{
1751 struct perf_callchain_entry *entry;
1752
1753 if (perf_guest_cbs && perf_guest_cbs->is_in_guest()) {
1754 /* TODO: We don't support guest os callchain now */
1755 return NULL;
1756 }
1757
1758 if (in_nmi())
1759 entry = &__get_cpu_var(pmc_nmi_entry);
1760 else
1761 entry = &__get_cpu_var(pmc_irq_entry);
1762
1763 entry->nr = 0;
1764
1765 perf_do_callchain(regs, entry);
1766
1767 return entry;
1768}
1769
1770unsigned long perf_instruction_pointer(struct pt_regs *regs) 1757unsigned long perf_instruction_pointer(struct pt_regs *regs)
1771{ 1758{
1772 unsigned long ip; 1759 unsigned long ip;
diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c
index ee05c90012d2..c8f5c088cad1 100644
--- a/arch/x86/kernel/cpu/perf_event_intel.c
+++ b/arch/x86/kernel/cpu/perf_event_intel.c
@@ -713,18 +713,18 @@ static int intel_pmu_handle_irq(struct pt_regs *regs)
713 struct cpu_hw_events *cpuc; 713 struct cpu_hw_events *cpuc;
714 int bit, loops; 714 int bit, loops;
715 u64 status; 715 u64 status;
716 int handled = 0; 716 int handled;
717 717
718 perf_sample_data_init(&data, 0); 718 perf_sample_data_init(&data, 0);
719 719
720 cpuc = &__get_cpu_var(cpu_hw_events); 720 cpuc = &__get_cpu_var(cpu_hw_events);
721 721
722 intel_pmu_disable_all(); 722 intel_pmu_disable_all();
723 intel_pmu_drain_bts_buffer(); 723 handled = intel_pmu_drain_bts_buffer();
724 status = intel_pmu_get_status(); 724 status = intel_pmu_get_status();
725 if (!status) { 725 if (!status) {
726 intel_pmu_enable_all(0); 726 intel_pmu_enable_all(0);
727 return 0; 727 return handled;
728 } 728 }
729 729
730 loops = 0; 730 loops = 0;
@@ -763,7 +763,7 @@ again:
763 data.period = event->hw.last_period; 763 data.period = event->hw.last_period;
764 764
765 if (perf_event_overflow(event, 1, &data, regs)) 765 if (perf_event_overflow(event, 1, &data, regs))
766 x86_pmu_stop(event); 766 x86_pmu_stop(event, 0);
767 } 767 }
768 768
769 /* 769 /*
diff --git a/arch/x86/kernel/cpu/perf_event_intel_ds.c b/arch/x86/kernel/cpu/perf_event_intel_ds.c
index 18018d1311cd..4977f9c400e5 100644
--- a/arch/x86/kernel/cpu/perf_event_intel_ds.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_ds.c
@@ -214,7 +214,7 @@ static void intel_pmu_disable_bts(void)
214 update_debugctlmsr(debugctlmsr); 214 update_debugctlmsr(debugctlmsr);
215} 215}
216 216
217static void intel_pmu_drain_bts_buffer(void) 217static int intel_pmu_drain_bts_buffer(void)
218{ 218{
219 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); 219 struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
220 struct debug_store *ds = cpuc->ds; 220 struct debug_store *ds = cpuc->ds;
@@ -231,16 +231,16 @@ static void intel_pmu_drain_bts_buffer(void)
231 struct pt_regs regs; 231 struct pt_regs regs;
232 232
233 if (!event) 233 if (!event)
234 return; 234 return 0;
235 235
236 if (!ds) 236 if (!ds)
237 return; 237 return 0;
238 238
239 at = (struct bts_record *)(unsigned long)ds->bts_buffer_base; 239 at = (struct bts_record *)(unsigned long)ds->bts_buffer_base;
240 top = (struct bts_record *)(unsigned long)ds->bts_index; 240 top = (struct bts_record *)(unsigned long)ds->bts_index;
241 241
242 if (top <= at) 242 if (top <= at)
243 return; 243 return 0;
244 244
245 ds->bts_index = ds->bts_buffer_base; 245 ds->bts_index = ds->bts_buffer_base;
246 246
@@ -256,7 +256,7 @@ static void intel_pmu_drain_bts_buffer(void)
256 perf_prepare_sample(&header, &data, event, &regs); 256 perf_prepare_sample(&header, &data, event, &regs);
257 257
258 if (perf_output_begin(&handle, event, header.size * (top - at), 1, 1)) 258 if (perf_output_begin(&handle, event, header.size * (top - at), 1, 1))
259 return; 259 return 1;
260 260
261 for (; at < top; at++) { 261 for (; at < top; at++) {
262 data.ip = at->from; 262 data.ip = at->from;
@@ -270,6 +270,7 @@ static void intel_pmu_drain_bts_buffer(void)
270 /* There's new data available. */ 270 /* There's new data available. */
271 event->hw.interrupts++; 271 event->hw.interrupts++;
272 event->pending_kill = POLL_IN; 272 event->pending_kill = POLL_IN;
273 return 1;
273} 274}
274 275
275/* 276/*
@@ -491,7 +492,7 @@ static void __intel_pmu_pebs_event(struct perf_event *event,
491 regs.flags &= ~PERF_EFLAGS_EXACT; 492 regs.flags &= ~PERF_EFLAGS_EXACT;
492 493
493 if (perf_event_overflow(event, 1, &data, &regs)) 494 if (perf_event_overflow(event, 1, &data, &regs))
494 x86_pmu_stop(event); 495 x86_pmu_stop(event, 0);
495} 496}
496 497
497static void intel_pmu_drain_pebs_core(struct pt_regs *iregs) 498static void intel_pmu_drain_pebs_core(struct pt_regs *iregs)
diff --git a/arch/x86/kernel/cpu/perf_event_p4.c b/arch/x86/kernel/cpu/perf_event_p4.c
index b560db3305be..c70c878ee02a 100644
--- a/arch/x86/kernel/cpu/perf_event_p4.c
+++ b/arch/x86/kernel/cpu/perf_event_p4.c
@@ -18,6 +18,8 @@
18struct p4_event_bind { 18struct p4_event_bind {
19 unsigned int opcode; /* Event code and ESCR selector */ 19 unsigned int opcode; /* Event code and ESCR selector */
20 unsigned int escr_msr[2]; /* ESCR MSR for this event */ 20 unsigned int escr_msr[2]; /* ESCR MSR for this event */
21 unsigned int escr_emask; /* valid ESCR EventMask bits */
22 unsigned int shared; /* event is shared across threads */
21 char cntr[2][P4_CNTR_LIMIT]; /* counter index (offset), -1 on abscence */ 23 char cntr[2][P4_CNTR_LIMIT]; /* counter index (offset), -1 on abscence */
22}; 24};
23 25
@@ -66,231 +68,435 @@ static struct p4_event_bind p4_event_bind_map[] = {
66 [P4_EVENT_TC_DELIVER_MODE] = { 68 [P4_EVENT_TC_DELIVER_MODE] = {
67 .opcode = P4_OPCODE(P4_EVENT_TC_DELIVER_MODE), 69 .opcode = P4_OPCODE(P4_EVENT_TC_DELIVER_MODE),
68 .escr_msr = { MSR_P4_TC_ESCR0, MSR_P4_TC_ESCR1 }, 70 .escr_msr = { MSR_P4_TC_ESCR0, MSR_P4_TC_ESCR1 },
71 .escr_emask =
72 P4_ESCR_EMASK_BIT(P4_EVENT_TC_DELIVER_MODE, DD) |
73 P4_ESCR_EMASK_BIT(P4_EVENT_TC_DELIVER_MODE, DB) |
74 P4_ESCR_EMASK_BIT(P4_EVENT_TC_DELIVER_MODE, DI) |
75 P4_ESCR_EMASK_BIT(P4_EVENT_TC_DELIVER_MODE, BD) |
76 P4_ESCR_EMASK_BIT(P4_EVENT_TC_DELIVER_MODE, BB) |
77 P4_ESCR_EMASK_BIT(P4_EVENT_TC_DELIVER_MODE, BI) |
78 P4_ESCR_EMASK_BIT(P4_EVENT_TC_DELIVER_MODE, ID),
79 .shared = 1,
69 .cntr = { {4, 5, -1}, {6, 7, -1} }, 80 .cntr = { {4, 5, -1}, {6, 7, -1} },
70 }, 81 },
71 [P4_EVENT_BPU_FETCH_REQUEST] = { 82 [P4_EVENT_BPU_FETCH_REQUEST] = {
72 .opcode = P4_OPCODE(P4_EVENT_BPU_FETCH_REQUEST), 83 .opcode = P4_OPCODE(P4_EVENT_BPU_FETCH_REQUEST),
73 .escr_msr = { MSR_P4_BPU_ESCR0, MSR_P4_BPU_ESCR1 }, 84 .escr_msr = { MSR_P4_BPU_ESCR0, MSR_P4_BPU_ESCR1 },
85 .escr_emask =
86 P4_ESCR_EMASK_BIT(P4_EVENT_BPU_FETCH_REQUEST, TCMISS),
74 .cntr = { {0, -1, -1}, {2, -1, -1} }, 87 .cntr = { {0, -1, -1}, {2, -1, -1} },
75 }, 88 },
76 [P4_EVENT_ITLB_REFERENCE] = { 89 [P4_EVENT_ITLB_REFERENCE] = {
77 .opcode = P4_OPCODE(P4_EVENT_ITLB_REFERENCE), 90 .opcode = P4_OPCODE(P4_EVENT_ITLB_REFERENCE),
78 .escr_msr = { MSR_P4_ITLB_ESCR0, MSR_P4_ITLB_ESCR1 }, 91 .escr_msr = { MSR_P4_ITLB_ESCR0, MSR_P4_ITLB_ESCR1 },
92 .escr_emask =
93 P4_ESCR_EMASK_BIT(P4_EVENT_ITLB_REFERENCE, HIT) |
94 P4_ESCR_EMASK_BIT(P4_EVENT_ITLB_REFERENCE, MISS) |
95 P4_ESCR_EMASK_BIT(P4_EVENT_ITLB_REFERENCE, HIT_UK),
79 .cntr = { {0, -1, -1}, {2, -1, -1} }, 96 .cntr = { {0, -1, -1}, {2, -1, -1} },
80 }, 97 },
81 [P4_EVENT_MEMORY_CANCEL] = { 98 [P4_EVENT_MEMORY_CANCEL] = {
82 .opcode = P4_OPCODE(P4_EVENT_MEMORY_CANCEL), 99 .opcode = P4_OPCODE(P4_EVENT_MEMORY_CANCEL),
83 .escr_msr = { MSR_P4_DAC_ESCR0, MSR_P4_DAC_ESCR1 }, 100 .escr_msr = { MSR_P4_DAC_ESCR0, MSR_P4_DAC_ESCR1 },
101 .escr_emask =
102 P4_ESCR_EMASK_BIT(P4_EVENT_MEMORY_CANCEL, ST_RB_FULL) |
103 P4_ESCR_EMASK_BIT(P4_EVENT_MEMORY_CANCEL, 64K_CONF),
84 .cntr = { {8, 9, -1}, {10, 11, -1} }, 104 .cntr = { {8, 9, -1}, {10, 11, -1} },
85 }, 105 },
86 [P4_EVENT_MEMORY_COMPLETE] = { 106 [P4_EVENT_MEMORY_COMPLETE] = {
87 .opcode = P4_OPCODE(P4_EVENT_MEMORY_COMPLETE), 107 .opcode = P4_OPCODE(P4_EVENT_MEMORY_COMPLETE),
88 .escr_msr = { MSR_P4_SAAT_ESCR0 , MSR_P4_SAAT_ESCR1 }, 108 .escr_msr = { MSR_P4_SAAT_ESCR0 , MSR_P4_SAAT_ESCR1 },
109 .escr_emask =
110 P4_ESCR_EMASK_BIT(P4_EVENT_MEMORY_COMPLETE, LSC) |
111 P4_ESCR_EMASK_BIT(P4_EVENT_MEMORY_COMPLETE, SSC),
89 .cntr = { {8, 9, -1}, {10, 11, -1} }, 112 .cntr = { {8, 9, -1}, {10, 11, -1} },
90 }, 113 },
91 [P4_EVENT_LOAD_PORT_REPLAY] = { 114 [P4_EVENT_LOAD_PORT_REPLAY] = {
92 .opcode = P4_OPCODE(P4_EVENT_LOAD_PORT_REPLAY), 115 .opcode = P4_OPCODE(P4_EVENT_LOAD_PORT_REPLAY),
93 .escr_msr = { MSR_P4_SAAT_ESCR0, MSR_P4_SAAT_ESCR1 }, 116 .escr_msr = { MSR_P4_SAAT_ESCR0, MSR_P4_SAAT_ESCR1 },
117 .escr_emask =
118 P4_ESCR_EMASK_BIT(P4_EVENT_LOAD_PORT_REPLAY, SPLIT_LD),
94 .cntr = { {8, 9, -1}, {10, 11, -1} }, 119 .cntr = { {8, 9, -1}, {10, 11, -1} },
95 }, 120 },
96 [P4_EVENT_STORE_PORT_REPLAY] = { 121 [P4_EVENT_STORE_PORT_REPLAY] = {
97 .opcode = P4_OPCODE(P4_EVENT_STORE_PORT_REPLAY), 122 .opcode = P4_OPCODE(P4_EVENT_STORE_PORT_REPLAY),
98 .escr_msr = { MSR_P4_SAAT_ESCR0 , MSR_P4_SAAT_ESCR1 }, 123 .escr_msr = { MSR_P4_SAAT_ESCR0 , MSR_P4_SAAT_ESCR1 },
124 .escr_emask =
125 P4_ESCR_EMASK_BIT(P4_EVENT_STORE_PORT_REPLAY, SPLIT_ST),
99 .cntr = { {8, 9, -1}, {10, 11, -1} }, 126 .cntr = { {8, 9, -1}, {10, 11, -1} },
100 }, 127 },
101 [P4_EVENT_MOB_LOAD_REPLAY] = { 128 [P4_EVENT_MOB_LOAD_REPLAY] = {
102 .opcode = P4_OPCODE(P4_EVENT_MOB_LOAD_REPLAY), 129 .opcode = P4_OPCODE(P4_EVENT_MOB_LOAD_REPLAY),
103 .escr_msr = { MSR_P4_MOB_ESCR0, MSR_P4_MOB_ESCR1 }, 130 .escr_msr = { MSR_P4_MOB_ESCR0, MSR_P4_MOB_ESCR1 },
131 .escr_emask =
132 P4_ESCR_EMASK_BIT(P4_EVENT_MOB_LOAD_REPLAY, NO_STA) |
133 P4_ESCR_EMASK_BIT(P4_EVENT_MOB_LOAD_REPLAY, NO_STD) |
134 P4_ESCR_EMASK_BIT(P4_EVENT_MOB_LOAD_REPLAY, PARTIAL_DATA) |
135 P4_ESCR_EMASK_BIT(P4_EVENT_MOB_LOAD_REPLAY, UNALGN_ADDR),
104 .cntr = { {0, -1, -1}, {2, -1, -1} }, 136 .cntr = { {0, -1, -1}, {2, -1, -1} },
105 }, 137 },
106 [P4_EVENT_PAGE_WALK_TYPE] = { 138 [P4_EVENT_PAGE_WALK_TYPE] = {
107 .opcode = P4_OPCODE(P4_EVENT_PAGE_WALK_TYPE), 139 .opcode = P4_OPCODE(P4_EVENT_PAGE_WALK_TYPE),
108 .escr_msr = { MSR_P4_PMH_ESCR0, MSR_P4_PMH_ESCR1 }, 140 .escr_msr = { MSR_P4_PMH_ESCR0, MSR_P4_PMH_ESCR1 },
141 .escr_emask =
142 P4_ESCR_EMASK_BIT(P4_EVENT_PAGE_WALK_TYPE, DTMISS) |
143 P4_ESCR_EMASK_BIT(P4_EVENT_PAGE_WALK_TYPE, ITMISS),
144 .shared = 1,
109 .cntr = { {0, -1, -1}, {2, -1, -1} }, 145 .cntr = { {0, -1, -1}, {2, -1, -1} },
110 }, 146 },
111 [P4_EVENT_BSQ_CACHE_REFERENCE] = { 147 [P4_EVENT_BSQ_CACHE_REFERENCE] = {
112 .opcode = P4_OPCODE(P4_EVENT_BSQ_CACHE_REFERENCE), 148 .opcode = P4_OPCODE(P4_EVENT_BSQ_CACHE_REFERENCE),
113 .escr_msr = { MSR_P4_BSU_ESCR0, MSR_P4_BSU_ESCR1 }, 149 .escr_msr = { MSR_P4_BSU_ESCR0, MSR_P4_BSU_ESCR1 },
150 .escr_emask =
151 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITS) |
152 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITE) |
153 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_HITM) |
154 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITS) |
155 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITE) |
156 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_HITM) |
157 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_2ndL_MISS) |
158 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, RD_3rdL_MISS) |
159 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_CACHE_REFERENCE, WR_2ndL_MISS),
114 .cntr = { {0, -1, -1}, {2, -1, -1} }, 160 .cntr = { {0, -1, -1}, {2, -1, -1} },
115 }, 161 },
116 [P4_EVENT_IOQ_ALLOCATION] = { 162 [P4_EVENT_IOQ_ALLOCATION] = {
117 .opcode = P4_OPCODE(P4_EVENT_IOQ_ALLOCATION), 163 .opcode = P4_OPCODE(P4_EVENT_IOQ_ALLOCATION),
118 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 }, 164 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
165 .escr_emask =
166 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, DEFAULT) |
167 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, ALL_READ) |
168 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, ALL_WRITE) |
169 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, MEM_UC) |
170 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, MEM_WC) |
171 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, MEM_WT) |
172 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, MEM_WP) |
173 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, MEM_WB) |
174 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, OWN) |
175 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, OTHER) |
176 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ALLOCATION, PREFETCH),
119 .cntr = { {0, -1, -1}, {2, -1, -1} }, 177 .cntr = { {0, -1, -1}, {2, -1, -1} },
120 }, 178 },
121 [P4_EVENT_IOQ_ACTIVE_ENTRIES] = { /* shared ESCR */ 179 [P4_EVENT_IOQ_ACTIVE_ENTRIES] = { /* shared ESCR */
122 .opcode = P4_OPCODE(P4_EVENT_IOQ_ACTIVE_ENTRIES), 180 .opcode = P4_OPCODE(P4_EVENT_IOQ_ACTIVE_ENTRIES),
123 .escr_msr = { MSR_P4_FSB_ESCR1, MSR_P4_FSB_ESCR1 }, 181 .escr_msr = { MSR_P4_FSB_ESCR1, MSR_P4_FSB_ESCR1 },
182 .escr_emask =
183 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, DEFAULT) |
184 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, ALL_READ) |
185 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, ALL_WRITE) |
186 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, MEM_UC) |
187 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, MEM_WC) |
188 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, MEM_WT) |
189 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, MEM_WP) |
190 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, MEM_WB) |
191 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, OWN) |
192 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, OTHER) |
193 P4_ESCR_EMASK_BIT(P4_EVENT_IOQ_ACTIVE_ENTRIES, PREFETCH),
124 .cntr = { {2, -1, -1}, {3, -1, -1} }, 194 .cntr = { {2, -1, -1}, {3, -1, -1} },
125 }, 195 },
126 [P4_EVENT_FSB_DATA_ACTIVITY] = { 196 [P4_EVENT_FSB_DATA_ACTIVITY] = {
127 .opcode = P4_OPCODE(P4_EVENT_FSB_DATA_ACTIVITY), 197 .opcode = P4_OPCODE(P4_EVENT_FSB_DATA_ACTIVITY),
128 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 }, 198 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
199 .escr_emask =
200 P4_ESCR_EMASK_BIT(P4_EVENT_FSB_DATA_ACTIVITY, DRDY_DRV) |
201 P4_ESCR_EMASK_BIT(P4_EVENT_FSB_DATA_ACTIVITY, DRDY_OWN) |
202 P4_ESCR_EMASK_BIT(P4_EVENT_FSB_DATA_ACTIVITY, DRDY_OTHER) |
203 P4_ESCR_EMASK_BIT(P4_EVENT_FSB_DATA_ACTIVITY, DBSY_DRV) |
204 P4_ESCR_EMASK_BIT(P4_EVENT_FSB_DATA_ACTIVITY, DBSY_OWN) |
205 P4_ESCR_EMASK_BIT(P4_EVENT_FSB_DATA_ACTIVITY, DBSY_OTHER),
206 .shared = 1,
129 .cntr = { {0, -1, -1}, {2, -1, -1} }, 207 .cntr = { {0, -1, -1}, {2, -1, -1} },
130 }, 208 },
131 [P4_EVENT_BSQ_ALLOCATION] = { /* shared ESCR, broken CCCR1 */ 209 [P4_EVENT_BSQ_ALLOCATION] = { /* shared ESCR, broken CCCR1 */
132 .opcode = P4_OPCODE(P4_EVENT_BSQ_ALLOCATION), 210 .opcode = P4_OPCODE(P4_EVENT_BSQ_ALLOCATION),
133 .escr_msr = { MSR_P4_BSU_ESCR0, MSR_P4_BSU_ESCR0 }, 211 .escr_msr = { MSR_P4_BSU_ESCR0, MSR_P4_BSU_ESCR0 },
212 .escr_emask =
213 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, REQ_TYPE0) |
214 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, REQ_TYPE1) |
215 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, REQ_LEN0) |
216 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, REQ_LEN1) |
217 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, REQ_IO_TYPE) |
218 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, REQ_LOCK_TYPE) |
219 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, REQ_CACHE_TYPE) |
220 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, REQ_SPLIT_TYPE) |
221 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, REQ_DEM_TYPE) |
222 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, REQ_ORD_TYPE) |
223 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, MEM_TYPE0) |
224 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, MEM_TYPE1) |
225 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ALLOCATION, MEM_TYPE2),
134 .cntr = { {0, -1, -1}, {1, -1, -1} }, 226 .cntr = { {0, -1, -1}, {1, -1, -1} },
135 }, 227 },
136 [P4_EVENT_BSQ_ACTIVE_ENTRIES] = { /* shared ESCR */ 228 [P4_EVENT_BSQ_ACTIVE_ENTRIES] = { /* shared ESCR */
137 .opcode = P4_OPCODE(P4_EVENT_BSQ_ACTIVE_ENTRIES), 229 .opcode = P4_OPCODE(P4_EVENT_BSQ_ACTIVE_ENTRIES),
138 .escr_msr = { MSR_P4_BSU_ESCR1 , MSR_P4_BSU_ESCR1 }, 230 .escr_msr = { MSR_P4_BSU_ESCR1 , MSR_P4_BSU_ESCR1 },
231 .escr_emask =
232 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_TYPE0) |
233 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_TYPE1) |
234 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_LEN0) |
235 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_LEN1) |
236 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_IO_TYPE) |
237 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_LOCK_TYPE) |
238 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_CACHE_TYPE) |
239 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_SPLIT_TYPE) |
240 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_DEM_TYPE) |
241 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, REQ_ORD_TYPE) |
242 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, MEM_TYPE0) |
243 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, MEM_TYPE1) |
244 P4_ESCR_EMASK_BIT(P4_EVENT_BSQ_ACTIVE_ENTRIES, MEM_TYPE2),
139 .cntr = { {2, -1, -1}, {3, -1, -1} }, 245 .cntr = { {2, -1, -1}, {3, -1, -1} },
140 }, 246 },
141 [P4_EVENT_SSE_INPUT_ASSIST] = { 247 [P4_EVENT_SSE_INPUT_ASSIST] = {
142 .opcode = P4_OPCODE(P4_EVENT_SSE_INPUT_ASSIST), 248 .opcode = P4_OPCODE(P4_EVENT_SSE_INPUT_ASSIST),
143 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 }, 249 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
250 .escr_emask =
251 P4_ESCR_EMASK_BIT(P4_EVENT_SSE_INPUT_ASSIST, ALL),
252 .shared = 1,
144 .cntr = { {8, 9, -1}, {10, 11, -1} }, 253 .cntr = { {8, 9, -1}, {10, 11, -1} },
145 }, 254 },
146 [P4_EVENT_PACKED_SP_UOP] = { 255 [P4_EVENT_PACKED_SP_UOP] = {
147 .opcode = P4_OPCODE(P4_EVENT_PACKED_SP_UOP), 256 .opcode = P4_OPCODE(P4_EVENT_PACKED_SP_UOP),
148 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 }, 257 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
258 .escr_emask =
259 P4_ESCR_EMASK_BIT(P4_EVENT_PACKED_SP_UOP, ALL),
260 .shared = 1,
149 .cntr = { {8, 9, -1}, {10, 11, -1} }, 261 .cntr = { {8, 9, -1}, {10, 11, -1} },
150 }, 262 },
151 [P4_EVENT_PACKED_DP_UOP] = { 263 [P4_EVENT_PACKED_DP_UOP] = {
152 .opcode = P4_OPCODE(P4_EVENT_PACKED_DP_UOP), 264 .opcode = P4_OPCODE(P4_EVENT_PACKED_DP_UOP),
153 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 }, 265 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
266 .escr_emask =
267 P4_ESCR_EMASK_BIT(P4_EVENT_PACKED_DP_UOP, ALL),
268 .shared = 1,
154 .cntr = { {8, 9, -1}, {10, 11, -1} }, 269 .cntr = { {8, 9, -1}, {10, 11, -1} },
155 }, 270 },
156 [P4_EVENT_SCALAR_SP_UOP] = { 271 [P4_EVENT_SCALAR_SP_UOP] = {
157 .opcode = P4_OPCODE(P4_EVENT_SCALAR_SP_UOP), 272 .opcode = P4_OPCODE(P4_EVENT_SCALAR_SP_UOP),
158 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 }, 273 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
274 .escr_emask =
275 P4_ESCR_EMASK_BIT(P4_EVENT_SCALAR_SP_UOP, ALL),
276 .shared = 1,
159 .cntr = { {8, 9, -1}, {10, 11, -1} }, 277 .cntr = { {8, 9, -1}, {10, 11, -1} },
160 }, 278 },
161 [P4_EVENT_SCALAR_DP_UOP] = { 279 [P4_EVENT_SCALAR_DP_UOP] = {
162 .opcode = P4_OPCODE(P4_EVENT_SCALAR_DP_UOP), 280 .opcode = P4_OPCODE(P4_EVENT_SCALAR_DP_UOP),
163 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 }, 281 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
282 .escr_emask =
283 P4_ESCR_EMASK_BIT(P4_EVENT_SCALAR_DP_UOP, ALL),
284 .shared = 1,
164 .cntr = { {8, 9, -1}, {10, 11, -1} }, 285 .cntr = { {8, 9, -1}, {10, 11, -1} },
165 }, 286 },
166 [P4_EVENT_64BIT_MMX_UOP] = { 287 [P4_EVENT_64BIT_MMX_UOP] = {
167 .opcode = P4_OPCODE(P4_EVENT_64BIT_MMX_UOP), 288 .opcode = P4_OPCODE(P4_EVENT_64BIT_MMX_UOP),
168 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 }, 289 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
290 .escr_emask =
291 P4_ESCR_EMASK_BIT(P4_EVENT_64BIT_MMX_UOP, ALL),
292 .shared = 1,
169 .cntr = { {8, 9, -1}, {10, 11, -1} }, 293 .cntr = { {8, 9, -1}, {10, 11, -1} },
170 }, 294 },
171 [P4_EVENT_128BIT_MMX_UOP] = { 295 [P4_EVENT_128BIT_MMX_UOP] = {
172 .opcode = P4_OPCODE(P4_EVENT_128BIT_MMX_UOP), 296 .opcode = P4_OPCODE(P4_EVENT_128BIT_MMX_UOP),
173 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 }, 297 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
298 .escr_emask =
299 P4_ESCR_EMASK_BIT(P4_EVENT_128BIT_MMX_UOP, ALL),
300 .shared = 1,
174 .cntr = { {8, 9, -1}, {10, 11, -1} }, 301 .cntr = { {8, 9, -1}, {10, 11, -1} },
175 }, 302 },
176 [P4_EVENT_X87_FP_UOP] = { 303 [P4_EVENT_X87_FP_UOP] = {
177 .opcode = P4_OPCODE(P4_EVENT_X87_FP_UOP), 304 .opcode = P4_OPCODE(P4_EVENT_X87_FP_UOP),
178 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 }, 305 .escr_msr = { MSR_P4_FIRM_ESCR0, MSR_P4_FIRM_ESCR1 },
306 .escr_emask =
307 P4_ESCR_EMASK_BIT(P4_EVENT_X87_FP_UOP, ALL),
308 .shared = 1,
179 .cntr = { {8, 9, -1}, {10, 11, -1} }, 309 .cntr = { {8, 9, -1}, {10, 11, -1} },
180 }, 310 },
181 [P4_EVENT_TC_MISC] = { 311 [P4_EVENT_TC_MISC] = {
182 .opcode = P4_OPCODE(P4_EVENT_TC_MISC), 312 .opcode = P4_OPCODE(P4_EVENT_TC_MISC),
183 .escr_msr = { MSR_P4_TC_ESCR0, MSR_P4_TC_ESCR1 }, 313 .escr_msr = { MSR_P4_TC_ESCR0, MSR_P4_TC_ESCR1 },
314 .escr_emask =
315 P4_ESCR_EMASK_BIT(P4_EVENT_TC_MISC, FLUSH),
184 .cntr = { {4, 5, -1}, {6, 7, -1} }, 316 .cntr = { {4, 5, -1}, {6, 7, -1} },
185 }, 317 },
186 [P4_EVENT_GLOBAL_POWER_EVENTS] = { 318 [P4_EVENT_GLOBAL_POWER_EVENTS] = {
187 .opcode = P4_OPCODE(P4_EVENT_GLOBAL_POWER_EVENTS), 319 .opcode = P4_OPCODE(P4_EVENT_GLOBAL_POWER_EVENTS),
188 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 }, 320 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
321 .escr_emask =
322 P4_ESCR_EMASK_BIT(P4_EVENT_GLOBAL_POWER_EVENTS, RUNNING),
189 .cntr = { {0, -1, -1}, {2, -1, -1} }, 323 .cntr = { {0, -1, -1}, {2, -1, -1} },
190 }, 324 },
191 [P4_EVENT_TC_MS_XFER] = { 325 [P4_EVENT_TC_MS_XFER] = {
192 .opcode = P4_OPCODE(P4_EVENT_TC_MS_XFER), 326 .opcode = P4_OPCODE(P4_EVENT_TC_MS_XFER),
193 .escr_msr = { MSR_P4_MS_ESCR0, MSR_P4_MS_ESCR1 }, 327 .escr_msr = { MSR_P4_MS_ESCR0, MSR_P4_MS_ESCR1 },
328 .escr_emask =
329 P4_ESCR_EMASK_BIT(P4_EVENT_TC_MS_XFER, CISC),
194 .cntr = { {4, 5, -1}, {6, 7, -1} }, 330 .cntr = { {4, 5, -1}, {6, 7, -1} },
195 }, 331 },
196 [P4_EVENT_UOP_QUEUE_WRITES] = { 332 [P4_EVENT_UOP_QUEUE_WRITES] = {
197 .opcode = P4_OPCODE(P4_EVENT_UOP_QUEUE_WRITES), 333 .opcode = P4_OPCODE(P4_EVENT_UOP_QUEUE_WRITES),
198 .escr_msr = { MSR_P4_MS_ESCR0, MSR_P4_MS_ESCR1 }, 334 .escr_msr = { MSR_P4_MS_ESCR0, MSR_P4_MS_ESCR1 },
335 .escr_emask =
336 P4_ESCR_EMASK_BIT(P4_EVENT_UOP_QUEUE_WRITES, FROM_TC_BUILD) |
337 P4_ESCR_EMASK_BIT(P4_EVENT_UOP_QUEUE_WRITES, FROM_TC_DELIVER) |
338 P4_ESCR_EMASK_BIT(P4_EVENT_UOP_QUEUE_WRITES, FROM_ROM),
199 .cntr = { {4, 5, -1}, {6, 7, -1} }, 339 .cntr = { {4, 5, -1}, {6, 7, -1} },
200 }, 340 },
201 [P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE] = { 341 [P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE] = {
202 .opcode = P4_OPCODE(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE), 342 .opcode = P4_OPCODE(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE),
203 .escr_msr = { MSR_P4_TBPU_ESCR0 , MSR_P4_TBPU_ESCR0 }, 343 .escr_msr = { MSR_P4_TBPU_ESCR0 , MSR_P4_TBPU_ESCR0 },
344 .escr_emask =
345 P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE, CONDITIONAL) |
346 P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE, CALL) |
347 P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE, RETURN) |
348 P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_MISPRED_BRANCH_TYPE, INDIRECT),
204 .cntr = { {4, 5, -1}, {6, 7, -1} }, 349 .cntr = { {4, 5, -1}, {6, 7, -1} },
205 }, 350 },
206 [P4_EVENT_RETIRED_BRANCH_TYPE] = { 351 [P4_EVENT_RETIRED_BRANCH_TYPE] = {
207 .opcode = P4_OPCODE(P4_EVENT_RETIRED_BRANCH_TYPE), 352 .opcode = P4_OPCODE(P4_EVENT_RETIRED_BRANCH_TYPE),
208 .escr_msr = { MSR_P4_TBPU_ESCR0 , MSR_P4_TBPU_ESCR1 }, 353 .escr_msr = { MSR_P4_TBPU_ESCR0 , MSR_P4_TBPU_ESCR1 },
354 .escr_emask =
355 P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, CONDITIONAL) |
356 P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, CALL) |
357 P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, RETURN) |
358 P4_ESCR_EMASK_BIT(P4_EVENT_RETIRED_BRANCH_TYPE, INDIRECT),
209 .cntr = { {4, 5, -1}, {6, 7, -1} }, 359 .cntr = { {4, 5, -1}, {6, 7, -1} },
210 }, 360 },
211 [P4_EVENT_RESOURCE_STALL] = { 361 [P4_EVENT_RESOURCE_STALL] = {
212 .opcode = P4_OPCODE(P4_EVENT_RESOURCE_STALL), 362 .opcode = P4_OPCODE(P4_EVENT_RESOURCE_STALL),
213 .escr_msr = { MSR_P4_ALF_ESCR0, MSR_P4_ALF_ESCR1 }, 363 .escr_msr = { MSR_P4_ALF_ESCR0, MSR_P4_ALF_ESCR1 },
364 .escr_emask =
365 P4_ESCR_EMASK_BIT(P4_EVENT_RESOURCE_STALL, SBFULL),
214 .cntr = { {12, 13, 16}, {14, 15, 17} }, 366 .cntr = { {12, 13, 16}, {14, 15, 17} },
215 }, 367 },
216 [P4_EVENT_WC_BUFFER] = { 368 [P4_EVENT_WC_BUFFER] = {
217 .opcode = P4_OPCODE(P4_EVENT_WC_BUFFER), 369 .opcode = P4_OPCODE(P4_EVENT_WC_BUFFER),
218 .escr_msr = { MSR_P4_DAC_ESCR0, MSR_P4_DAC_ESCR1 }, 370 .escr_msr = { MSR_P4_DAC_ESCR0, MSR_P4_DAC_ESCR1 },
371 .escr_emask =
372 P4_ESCR_EMASK_BIT(P4_EVENT_WC_BUFFER, WCB_EVICTS) |
373 P4_ESCR_EMASK_BIT(P4_EVENT_WC_BUFFER, WCB_FULL_EVICTS),
374 .shared = 1,
219 .cntr = { {8, 9, -1}, {10, 11, -1} }, 375 .cntr = { {8, 9, -1}, {10, 11, -1} },
220 }, 376 },
221 [P4_EVENT_B2B_CYCLES] = { 377 [P4_EVENT_B2B_CYCLES] = {
222 .opcode = P4_OPCODE(P4_EVENT_B2B_CYCLES), 378 .opcode = P4_OPCODE(P4_EVENT_B2B_CYCLES),
223 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 }, 379 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
380 .escr_emask = 0,
224 .cntr = { {0, -1, -1}, {2, -1, -1} }, 381 .cntr = { {0, -1, -1}, {2, -1, -1} },
225 }, 382 },
226 [P4_EVENT_BNR] = { 383 [P4_EVENT_BNR] = {
227 .opcode = P4_OPCODE(P4_EVENT_BNR), 384 .opcode = P4_OPCODE(P4_EVENT_BNR),
228 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 }, 385 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
386 .escr_emask = 0,
229 .cntr = { {0, -1, -1}, {2, -1, -1} }, 387 .cntr = { {0, -1, -1}, {2, -1, -1} },
230 }, 388 },
231 [P4_EVENT_SNOOP] = { 389 [P4_EVENT_SNOOP] = {
232 .opcode = P4_OPCODE(P4_EVENT_SNOOP), 390 .opcode = P4_OPCODE(P4_EVENT_SNOOP),
233 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 }, 391 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
392 .escr_emask = 0,
234 .cntr = { {0, -1, -1}, {2, -1, -1} }, 393 .cntr = { {0, -1, -1}, {2, -1, -1} },
235 }, 394 },
236 [P4_EVENT_RESPONSE] = { 395 [P4_EVENT_RESPONSE] = {
237 .opcode = P4_OPCODE(P4_EVENT_RESPONSE), 396 .opcode = P4_OPCODE(P4_EVENT_RESPONSE),
238 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 }, 397 .escr_msr = { MSR_P4_FSB_ESCR0, MSR_P4_FSB_ESCR1 },
398 .escr_emask = 0,
239 .cntr = { {0, -1, -1}, {2, -1, -1} }, 399 .cntr = { {0, -1, -1}, {2, -1, -1} },
240 }, 400 },
241 [P4_EVENT_FRONT_END_EVENT] = { 401 [P4_EVENT_FRONT_END_EVENT] = {
242 .opcode = P4_OPCODE(P4_EVENT_FRONT_END_EVENT), 402 .opcode = P4_OPCODE(P4_EVENT_FRONT_END_EVENT),
243 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 }, 403 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
404 .escr_emask =
405 P4_ESCR_EMASK_BIT(P4_EVENT_FRONT_END_EVENT, NBOGUS) |
406 P4_ESCR_EMASK_BIT(P4_EVENT_FRONT_END_EVENT, BOGUS),
244 .cntr = { {12, 13, 16}, {14, 15, 17} }, 407 .cntr = { {12, 13, 16}, {14, 15, 17} },
245 }, 408 },
246 [P4_EVENT_EXECUTION_EVENT] = { 409 [P4_EVENT_EXECUTION_EVENT] = {
247 .opcode = P4_OPCODE(P4_EVENT_EXECUTION_EVENT), 410 .opcode = P4_OPCODE(P4_EVENT_EXECUTION_EVENT),
248 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 }, 411 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
412 .escr_emask =
413 P4_ESCR_EMASK_BIT(P4_EVENT_EXECUTION_EVENT, NBOGUS0) |
414 P4_ESCR_EMASK_BIT(P4_EVENT_EXECUTION_EVENT, NBOGUS1) |
415 P4_ESCR_EMASK_BIT(P4_EVENT_EXECUTION_EVENT, NBOGUS2) |
416 P4_ESCR_EMASK_BIT(P4_EVENT_EXECUTION_EVENT, NBOGUS3) |
417 P4_ESCR_EMASK_BIT(P4_EVENT_EXECUTION_EVENT, BOGUS0) |
418 P4_ESCR_EMASK_BIT(P4_EVENT_EXECUTION_EVENT, BOGUS1) |
419 P4_ESCR_EMASK_BIT(P4_EVENT_EXECUTION_EVENT, BOGUS2) |
420 P4_ESCR_EMASK_BIT(P4_EVENT_EXECUTION_EVENT, BOGUS3),
249 .cntr = { {12, 13, 16}, {14, 15, 17} }, 421 .cntr = { {12, 13, 16}, {14, 15, 17} },
250 }, 422 },
251 [P4_EVENT_REPLAY_EVENT] = { 423 [P4_EVENT_REPLAY_EVENT] = {
252 .opcode = P4_OPCODE(P4_EVENT_REPLAY_EVENT), 424 .opcode = P4_OPCODE(P4_EVENT_REPLAY_EVENT),
253 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 }, 425 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
426 .escr_emask =
427 P4_ESCR_EMASK_BIT(P4_EVENT_REPLAY_EVENT, NBOGUS) |
428 P4_ESCR_EMASK_BIT(P4_EVENT_REPLAY_EVENT, BOGUS),
254 .cntr = { {12, 13, 16}, {14, 15, 17} }, 429 .cntr = { {12, 13, 16}, {14, 15, 17} },
255 }, 430 },
256 [P4_EVENT_INSTR_RETIRED] = { 431 [P4_EVENT_INSTR_RETIRED] = {
257 .opcode = P4_OPCODE(P4_EVENT_INSTR_RETIRED), 432 .opcode = P4_OPCODE(P4_EVENT_INSTR_RETIRED),
258 .escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 }, 433 .escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
434 .escr_emask =
435 P4_ESCR_EMASK_BIT(P4_EVENT_INSTR_RETIRED, NBOGUSNTAG) |
436 P4_ESCR_EMASK_BIT(P4_EVENT_INSTR_RETIRED, NBOGUSTAG) |
437 P4_ESCR_EMASK_BIT(P4_EVENT_INSTR_RETIRED, BOGUSNTAG) |
438 P4_ESCR_EMASK_BIT(P4_EVENT_INSTR_RETIRED, BOGUSTAG),
259 .cntr = { {12, 13, 16}, {14, 15, 17} }, 439 .cntr = { {12, 13, 16}, {14, 15, 17} },
260 }, 440 },
261 [P4_EVENT_UOPS_RETIRED] = { 441 [P4_EVENT_UOPS_RETIRED] = {
262 .opcode = P4_OPCODE(P4_EVENT_UOPS_RETIRED), 442 .opcode = P4_OPCODE(P4_EVENT_UOPS_RETIRED),
263 .escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 }, 443 .escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
444 .escr_emask =
445 P4_ESCR_EMASK_BIT(P4_EVENT_UOPS_RETIRED, NBOGUS) |
446 P4_ESCR_EMASK_BIT(P4_EVENT_UOPS_RETIRED, BOGUS),
264 .cntr = { {12, 13, 16}, {14, 15, 17} }, 447 .cntr = { {12, 13, 16}, {14, 15, 17} },
265 }, 448 },
266 [P4_EVENT_UOP_TYPE] = { 449 [P4_EVENT_UOP_TYPE] = {
267 .opcode = P4_OPCODE(P4_EVENT_UOP_TYPE), 450 .opcode = P4_OPCODE(P4_EVENT_UOP_TYPE),
268 .escr_msr = { MSR_P4_RAT_ESCR0, MSR_P4_RAT_ESCR1 }, 451 .escr_msr = { MSR_P4_RAT_ESCR0, MSR_P4_RAT_ESCR1 },
452 .escr_emask =
453 P4_ESCR_EMASK_BIT(P4_EVENT_UOP_TYPE, TAGLOADS) |
454 P4_ESCR_EMASK_BIT(P4_EVENT_UOP_TYPE, TAGSTORES),
269 .cntr = { {12, 13, 16}, {14, 15, 17} }, 455 .cntr = { {12, 13, 16}, {14, 15, 17} },
270 }, 456 },
271 [P4_EVENT_BRANCH_RETIRED] = { 457 [P4_EVENT_BRANCH_RETIRED] = {
272 .opcode = P4_OPCODE(P4_EVENT_BRANCH_RETIRED), 458 .opcode = P4_OPCODE(P4_EVENT_BRANCH_RETIRED),
273 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 }, 459 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
460 .escr_emask =
461 P4_ESCR_EMASK_BIT(P4_EVENT_BRANCH_RETIRED, MMNP) |
462 P4_ESCR_EMASK_BIT(P4_EVENT_BRANCH_RETIRED, MMNM) |
463 P4_ESCR_EMASK_BIT(P4_EVENT_BRANCH_RETIRED, MMTP) |
464 P4_ESCR_EMASK_BIT(P4_EVENT_BRANCH_RETIRED, MMTM),
274 .cntr = { {12, 13, 16}, {14, 15, 17} }, 465 .cntr = { {12, 13, 16}, {14, 15, 17} },
275 }, 466 },
276 [P4_EVENT_MISPRED_BRANCH_RETIRED] = { 467 [P4_EVENT_MISPRED_BRANCH_RETIRED] = {
277 .opcode = P4_OPCODE(P4_EVENT_MISPRED_BRANCH_RETIRED), 468 .opcode = P4_OPCODE(P4_EVENT_MISPRED_BRANCH_RETIRED),
278 .escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 }, 469 .escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
470 .escr_emask =
471 P4_ESCR_EMASK_BIT(P4_EVENT_MISPRED_BRANCH_RETIRED, NBOGUS),
279 .cntr = { {12, 13, 16}, {14, 15, 17} }, 472 .cntr = { {12, 13, 16}, {14, 15, 17} },
280 }, 473 },
281 [P4_EVENT_X87_ASSIST] = { 474 [P4_EVENT_X87_ASSIST] = {
282 .opcode = P4_OPCODE(P4_EVENT_X87_ASSIST), 475 .opcode = P4_OPCODE(P4_EVENT_X87_ASSIST),
283 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 }, 476 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
477 .escr_emask =
478 P4_ESCR_EMASK_BIT(P4_EVENT_X87_ASSIST, FPSU) |
479 P4_ESCR_EMASK_BIT(P4_EVENT_X87_ASSIST, FPSO) |
480 P4_ESCR_EMASK_BIT(P4_EVENT_X87_ASSIST, POAO) |
481 P4_ESCR_EMASK_BIT(P4_EVENT_X87_ASSIST, POAU) |
482 P4_ESCR_EMASK_BIT(P4_EVENT_X87_ASSIST, PREA),
284 .cntr = { {12, 13, 16}, {14, 15, 17} }, 483 .cntr = { {12, 13, 16}, {14, 15, 17} },
285 }, 484 },
286 [P4_EVENT_MACHINE_CLEAR] = { 485 [P4_EVENT_MACHINE_CLEAR] = {
287 .opcode = P4_OPCODE(P4_EVENT_MACHINE_CLEAR), 486 .opcode = P4_OPCODE(P4_EVENT_MACHINE_CLEAR),
288 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 }, 487 .escr_msr = { MSR_P4_CRU_ESCR2, MSR_P4_CRU_ESCR3 },
488 .escr_emask =
489 P4_ESCR_EMASK_BIT(P4_EVENT_MACHINE_CLEAR, CLEAR) |
490 P4_ESCR_EMASK_BIT(P4_EVENT_MACHINE_CLEAR, MOCLEAR) |
491 P4_ESCR_EMASK_BIT(P4_EVENT_MACHINE_CLEAR, SMCLEAR),
289 .cntr = { {12, 13, 16}, {14, 15, 17} }, 492 .cntr = { {12, 13, 16}, {14, 15, 17} },
290 }, 493 },
291 [P4_EVENT_INSTR_COMPLETED] = { 494 [P4_EVENT_INSTR_COMPLETED] = {
292 .opcode = P4_OPCODE(P4_EVENT_INSTR_COMPLETED), 495 .opcode = P4_OPCODE(P4_EVENT_INSTR_COMPLETED),
293 .escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 }, 496 .escr_msr = { MSR_P4_CRU_ESCR0, MSR_P4_CRU_ESCR1 },
497 .escr_emask =
498 P4_ESCR_EMASK_BIT(P4_EVENT_INSTR_COMPLETED, NBOGUS) |
499 P4_ESCR_EMASK_BIT(P4_EVENT_INSTR_COMPLETED, BOGUS),
294 .cntr = { {12, 13, 16}, {14, 15, 17} }, 500 .cntr = { {12, 13, 16}, {14, 15, 17} },
295 }, 501 },
296}; 502};
@@ -428,29 +634,73 @@ static u64 p4_pmu_event_map(int hw_event)
428 return config; 634 return config;
429} 635}
430 636
637/* check cpu model specifics */
638static bool p4_event_match_cpu_model(unsigned int event_idx)
639{
640 /* INSTR_COMPLETED event only exist for model 3, 4, 6 (Prescott) */
641 if (event_idx == P4_EVENT_INSTR_COMPLETED) {
642 if (boot_cpu_data.x86_model != 3 &&
643 boot_cpu_data.x86_model != 4 &&
644 boot_cpu_data.x86_model != 6)
645 return false;
646 }
647
648 /*
649 * For info
650 * - IQ_ESCR0, IQ_ESCR1 only for models 1 and 2
651 */
652
653 return true;
654}
655
431static int p4_validate_raw_event(struct perf_event *event) 656static int p4_validate_raw_event(struct perf_event *event)
432{ 657{
433 unsigned int v; 658 unsigned int v, emask;
434 659
435 /* user data may have out-of-bound event index */ 660 /* User data may have out-of-bound event index */
436 v = p4_config_unpack_event(event->attr.config); 661 v = p4_config_unpack_event(event->attr.config);
437 if (v >= ARRAY_SIZE(p4_event_bind_map)) { 662 if (v >= ARRAY_SIZE(p4_event_bind_map))
438 pr_warning("P4 PMU: Unknown event code: %d\n", v); 663 return -EINVAL;
664
665 /* It may be unsupported: */
666 if (!p4_event_match_cpu_model(v))
439 return -EINVAL; 667 return -EINVAL;
668
669 /*
670 * NOTE: P4_CCCR_THREAD_ANY has not the same meaning as
671 * in Architectural Performance Monitoring, it means not
672 * on _which_ logical cpu to count but rather _when_, ie it
673 * depends on logical cpu state -- count event if one cpu active,
674 * none, both or any, so we just allow user to pass any value
675 * desired.
676 *
677 * In turn we always set Tx_OS/Tx_USR bits bound to logical
678 * cpu without their propagation to another cpu
679 */
680
681 /*
682 * if an event is shared accross the logical threads
683 * the user needs special permissions to be able to use it
684 */
685 if (p4_event_bind_map[v].shared) {
686 if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN))
687 return -EACCES;
440 } 688 }
441 689
690 /* ESCR EventMask bits may be invalid */
691 emask = p4_config_unpack_escr(event->attr.config) & P4_ESCR_EVENTMASK_MASK;
692 if (emask & ~p4_event_bind_map[v].escr_emask)
693 return -EINVAL;
694
442 /* 695 /*
443 * it may have some screwed PEBS bits 696 * it may have some invalid PEBS bits
444 */ 697 */
445 if (p4_config_pebs_has(event->attr.config, P4_PEBS_CONFIG_ENABLE)) { 698 if (p4_config_pebs_has(event->attr.config, P4_PEBS_CONFIG_ENABLE))
446 pr_warning("P4 PMU: PEBS are not supported yet\n");
447 return -EINVAL; 699 return -EINVAL;
448 } 700
449 v = p4_config_unpack_metric(event->attr.config); 701 v = p4_config_unpack_metric(event->attr.config);
450 if (v >= ARRAY_SIZE(p4_pebs_bind_map)) { 702 if (v >= ARRAY_SIZE(p4_pebs_bind_map))
451 pr_warning("P4 PMU: Unknown metric code: %d\n", v);
452 return -EINVAL; 703 return -EINVAL;
453 }
454 704
455 return 0; 705 return 0;
456} 706}
@@ -478,27 +728,21 @@ static int p4_hw_config(struct perf_event *event)
478 728
479 if (event->attr.type == PERF_TYPE_RAW) { 729 if (event->attr.type == PERF_TYPE_RAW) {
480 730
731 /*
732 * Clear bits we reserve to be managed by kernel itself
733 * and never allowed from a user space
734 */
735 event->attr.config &= P4_CONFIG_MASK;
736
481 rc = p4_validate_raw_event(event); 737 rc = p4_validate_raw_event(event);
482 if (rc) 738 if (rc)
483 goto out; 739 goto out;
484 740
485 /* 741 /*
486 * We don't control raw events so it's up to the caller
487 * to pass sane values (and we don't count the thread number
488 * on HT machine but allow HT-compatible specifics to be
489 * passed on)
490 *
491 * Note that for RAW events we allow user to use P4_CCCR_RESERVED 742 * Note that for RAW events we allow user to use P4_CCCR_RESERVED
492 * bits since we keep additional info here (for cache events and etc) 743 * bits since we keep additional info here (for cache events and etc)
493 *
494 * XXX: HT wide things should check perf_paranoid_cpu() &&
495 * CAP_SYS_ADMIN
496 */ 744 */
497 event->hw.config |= event->attr.config & 745 event->hw.config |= event->attr.config;
498 (p4_config_pack_escr(P4_ESCR_MASK_HT) |
499 p4_config_pack_cccr(P4_CCCR_MASK_HT | P4_CCCR_RESERVED));
500
501 event->hw.config &= ~P4_CCCR_FORCE_OVF;
502 } 746 }
503 747
504 rc = x86_setup_perfctr(event); 748 rc = x86_setup_perfctr(event);
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-27 09:02:18 -0500