/*
* Performance counter x86 architecture code
*
- * Copyright(C) 2008 Thomas Gleixner <tglx@linutronix.de>
- * Copyright(C) 2008 Red Hat, Inc., Ingo Molnar
- * Copyright(C) 2009 Jaswinder Singh Rajput
+ * Copyright (C) 2008 Thomas Gleixner <tglx@linutronix.de>
+ * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar
+ * Copyright (C) 2009 Jaswinder Singh Rajput
+ * Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter
+ * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
*
* For licencing details see kernel-base/COPYING
*/
#include <linux/module.h>
#include <linux/kdebug.h>
#include <linux/sched.h>
+#include <linux/uaccess.h>
#include <asm/apic.h>
+#include <asm/stacktrace.h>
+#include <asm/nmi.h>
-static bool perf_counters_initialized __read_mostly;
-
-/*
- * Number of (generic) HW counters:
- */
-static int nr_counters_generic __read_mostly;
static u64 perf_counter_mask __read_mostly;
-static u64 counter_value_mask __read_mostly;
-static int counter_value_bits __read_mostly;
-
-static int nr_counters_fixed __read_mostly;
struct cpu_hw_counters {
struct perf_counter *counters[X86_PMC_IDX_MAX];
- unsigned long used[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
- unsigned long interrupts;
- u64 throttle_ctrl;
+ unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
+ unsigned long interrupts;
int enabled;
};
/*
- * struct pmc_x86_ops - performance counter x86 ops
+ * struct x86_pmu - generic x86 pmu
*/
-struct pmc_x86_ops {
- u64 (*save_disable_all)(void);
- void (*restore_all)(u64);
- u64 (*get_status)(u64);
- void (*ack_status)(u64);
- void (*enable)(int, u64);
- void (*disable)(int, u64);
+struct x86_pmu {
+ const char *name;
+ int version;
+ int (*handle_irq)(struct pt_regs *);
+ void (*disable_all)(void);
+ void (*enable_all)(void);
+ void (*enable)(struct hw_perf_counter *, int);
+ void (*disable)(struct hw_perf_counter *, int);
unsigned eventsel;
unsigned perfctr;
u64 (*event_map)(int);
u64 (*raw_event)(u64);
int max_events;
+ int num_counters;
+ int num_counters_fixed;
+ int counter_bits;
+ u64 counter_mask;
+ u64 max_period;
+ u64 intel_ctrl;
};
-static struct pmc_x86_ops *pmc_ops;
+static struct x86_pmu x86_pmu __read_mostly;
static DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters) = {
.enabled = 1,
*/
static const u64 intel_perfmon_event_map[] =
{
- [PERF_COUNT_CPU_CYCLES] = 0x003c,
- [PERF_COUNT_INSTRUCTIONS] = 0x00c0,
- [PERF_COUNT_CACHE_REFERENCES] = 0x4f2e,
- [PERF_COUNT_CACHE_MISSES] = 0x412e,
- [PERF_COUNT_BRANCH_INSTRUCTIONS] = 0x00c4,
- [PERF_COUNT_BRANCH_MISSES] = 0x00c5,
- [PERF_COUNT_BUS_CYCLES] = 0x013c,
+ [PERF_COUNT_HW_CPU_CYCLES] = 0x003c,
+ [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0,
+ [PERF_COUNT_HW_CACHE_REFERENCES] = 0x4f2e,
+ [PERF_COUNT_HW_CACHE_MISSES] = 0x412e,
+ [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4,
+ [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5,
+ [PERF_COUNT_HW_BUS_CYCLES] = 0x013c,
};
-static u64 pmc_intel_event_map(int event)
+static u64 intel_pmu_event_map(int event)
{
return intel_perfmon_event_map[event];
}
-static u64 pmc_intel_raw_event(u64 event)
+/*
+ * Generalized hw caching related event table, filled
+ * in on a per model basis. A value of 0 means
+ * 'not supported', -1 means 'event makes no sense on
+ * this CPU', any other value means the raw event
+ * ID.
+ */
+
+#define C(x) PERF_COUNT_HW_CACHE_##x
+
+static u64 __read_mostly hw_cache_event_ids
+ [PERF_COUNT_HW_CACHE_MAX]
+ [PERF_COUNT_HW_CACHE_OP_MAX]
+ [PERF_COUNT_HW_CACHE_RESULT_MAX];
+
+static const u64 nehalem_hw_cache_event_ids
+ [PERF_COUNT_HW_CACHE_MAX]
+ [PERF_COUNT_HW_CACHE_OP_MAX]
+ [PERF_COUNT_HW_CACHE_RESULT_MAX] =
{
-#define CORE_EVNTSEL_EVENT_MASK 0x000000FF
-#define CORE_EVNTSEL_UNIT_MASK 0x0000FF00
-#define CORE_EVNTSEL_COUNTER_MASK 0xFF000000
+ [ C(L1D) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI */
+ [ C(RESULT_MISS) ] = 0x0140, /* L1D_CACHE_LD.I_STATE */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI */
+ [ C(RESULT_MISS) ] = 0x0141, /* L1D_CACHE_ST.I_STATE */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0x014e, /* L1D_PREFETCH.REQUESTS */
+ [ C(RESULT_MISS) ] = 0x024e, /* L1D_PREFETCH.MISS */
+ },
+ },
+ [ C(L1I ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */
+ [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0,
+ [ C(RESULT_MISS) ] = 0x0,
+ },
+ },
+ [ C(L2 ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0324, /* L2_RQSTS.LOADS */
+ [ C(RESULT_MISS) ] = 0x0224, /* L2_RQSTS.LD_MISS */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0c24, /* L2_RQSTS.RFOS */
+ [ C(RESULT_MISS) ] = 0x0824, /* L2_RQSTS.RFO_MISS */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0xc024, /* L2_RQSTS.PREFETCHES */
+ [ C(RESULT_MISS) ] = 0x8024, /* L2_RQSTS.PREFETCH_MISS */
+ },
+ },
+ [ C(DTLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI (alias) */
+ [ C(RESULT_MISS) ] = 0x0108, /* DTLB_LOAD_MISSES.ANY */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI (alias) */
+ [ C(RESULT_MISS) ] = 0x010c, /* MEM_STORE_RETIRED.DTLB_MISS */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0,
+ [ C(RESULT_MISS) ] = 0x0,
+ },
+ },
+ [ C(ITLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x01c0, /* INST_RETIRED.ANY_P */
+ [ C(RESULT_MISS) ] = 0x20c8, /* ITLB_MISS_RETIRED */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
+ [ C(BPU ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ALL_BRANCHES */
+ [ C(RESULT_MISS) ] = 0x03e8, /* BPU_CLEARS.ANY */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
+};
-#define CORE_EVNTSEL_MASK \
+static const u64 core2_hw_cache_event_ids
+ [PERF_COUNT_HW_CACHE_MAX]
+ [PERF_COUNT_HW_CACHE_OP_MAX]
+ [PERF_COUNT_HW_CACHE_RESULT_MAX] =
+{
+ [ C(L1D) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI */
+ [ C(RESULT_MISS) ] = 0x0140, /* L1D_CACHE_LD.I_STATE */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI */
+ [ C(RESULT_MISS) ] = 0x0141, /* L1D_CACHE_ST.I_STATE */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0x104e, /* L1D_PREFETCH.REQUESTS */
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(L1I ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0080, /* L1I.READS */
+ [ C(RESULT_MISS) ] = 0x0081, /* L1I.MISSES */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(L2 ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x4f29, /* L2_LD.MESI */
+ [ C(RESULT_MISS) ] = 0x4129, /* L2_LD.ISTATE */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0x4f2A, /* L2_ST.MESI */
+ [ C(RESULT_MISS) ] = 0x412A, /* L2_ST.ISTATE */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(DTLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI (alias) */
+ [ C(RESULT_MISS) ] = 0x0208, /* DTLB_MISSES.MISS_LD */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI (alias) */
+ [ C(RESULT_MISS) ] = 0x0808, /* DTLB_MISSES.MISS_ST */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(ITLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x00c0, /* INST_RETIRED.ANY_P */
+ [ C(RESULT_MISS) ] = 0x1282, /* ITLBMISSES */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
+ [ C(BPU ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ANY */
+ [ C(RESULT_MISS) ] = 0x00c5, /* BP_INST_RETIRED.MISPRED */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
+};
+
+static const u64 atom_hw_cache_event_ids
+ [PERF_COUNT_HW_CACHE_MAX]
+ [PERF_COUNT_HW_CACHE_OP_MAX]
+ [PERF_COUNT_HW_CACHE_RESULT_MAX] =
+{
+ [ C(L1D) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x2140, /* L1D_CACHE.LD */
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0x2240, /* L1D_CACHE.ST */
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(L1I ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */
+ [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(L2 ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x4f29, /* L2_LD.MESI */
+ [ C(RESULT_MISS) ] = 0x4129, /* L2_LD.ISTATE */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0x4f2A, /* L2_ST.MESI */
+ [ C(RESULT_MISS) ] = 0x412A, /* L2_ST.ISTATE */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(DTLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x2140, /* L1D_CACHE_LD.MESI (alias) */
+ [ C(RESULT_MISS) ] = 0x0508, /* DTLB_MISSES.MISS_LD */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0x2240, /* L1D_CACHE_ST.MESI (alias) */
+ [ C(RESULT_MISS) ] = 0x0608, /* DTLB_MISSES.MISS_ST */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(ITLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x00c0, /* INST_RETIRED.ANY_P */
+ [ C(RESULT_MISS) ] = 0x0282, /* ITLB.MISSES */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
+ [ C(BPU ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ANY */
+ [ C(RESULT_MISS) ] = 0x00c5, /* BP_INST_RETIRED.MISPRED */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
+};
+
+static u64 intel_pmu_raw_event(u64 event)
+{
+#define CORE_EVNTSEL_EVENT_MASK 0x000000FFULL
+#define CORE_EVNTSEL_UNIT_MASK 0x0000FF00ULL
+#define CORE_EVNTSEL_EDGE_MASK 0x00040000ULL
+#define CORE_EVNTSEL_INV_MASK 0x00800000ULL
+#define CORE_EVNTSEL_COUNTER_MASK 0xFF000000ULL
+
+#define CORE_EVNTSEL_MASK \
(CORE_EVNTSEL_EVENT_MASK | \
CORE_EVNTSEL_UNIT_MASK | \
+ CORE_EVNTSEL_EDGE_MASK | \
+ CORE_EVNTSEL_INV_MASK | \
CORE_EVNTSEL_COUNTER_MASK)
return event & CORE_EVNTSEL_MASK;
}
+static const u64 amd_0f_hw_cache_event_ids
+ [PERF_COUNT_HW_CACHE_MAX]
+ [PERF_COUNT_HW_CACHE_OP_MAX]
+ [PERF_COUNT_HW_CACHE_RESULT_MAX] =
+{
+ [ C(L1D) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(L1I ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0080, /* Instruction cache fetches */
+ [ C(RESULT_MISS) ] = 0x0081, /* Instruction cache misses */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(L2 ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(DTLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(ITLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0080, /* Instruction fecthes */
+ [ C(RESULT_MISS) ] = 0x0085, /* Instr. fetch ITLB misses */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
+ [ C(BPU ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x00c2, /* Retired Branch Instr. */
+ [ C(RESULT_MISS) ] = 0x00c3, /* Retired Mispredicted BI */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
+};
+
/*
* AMD Performance Monitor K7 and later.
*/
static const u64 amd_perfmon_event_map[] =
{
- [PERF_COUNT_CPU_CYCLES] = 0x0076,
- [PERF_COUNT_INSTRUCTIONS] = 0x00c0,
- [PERF_COUNT_CACHE_REFERENCES] = 0x0080,
- [PERF_COUNT_CACHE_MISSES] = 0x0081,
- [PERF_COUNT_BRANCH_INSTRUCTIONS] = 0x00c4,
- [PERF_COUNT_BRANCH_MISSES] = 0x00c5,
+ [PERF_COUNT_HW_CPU_CYCLES] = 0x0076,
+ [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0,
+ [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0080,
+ [PERF_COUNT_HW_CACHE_MISSES] = 0x0081,
+ [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4,
+ [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5,
};
-static u64 pmc_amd_event_map(int event)
+static u64 amd_pmu_event_map(int event)
{
return amd_perfmon_event_map[event];
}
-static u64 pmc_amd_raw_event(u64 event)
+static u64 amd_pmu_raw_event(u64 event)
{
-#define K7_EVNTSEL_EVENT_MASK 0x7000000FF
-#define K7_EVNTSEL_UNIT_MASK 0x00000FF00
-#define K7_EVNTSEL_COUNTER_MASK 0x0FF000000
+#define K7_EVNTSEL_EVENT_MASK 0x7000000FFULL
+#define K7_EVNTSEL_UNIT_MASK 0x00000FF00ULL
+#define K7_EVNTSEL_EDGE_MASK 0x000040000ULL
+#define K7_EVNTSEL_INV_MASK 0x000800000ULL
+#define K7_EVNTSEL_COUNTER_MASK 0x0FF000000ULL
#define K7_EVNTSEL_MASK \
(K7_EVNTSEL_EVENT_MASK | \
K7_EVNTSEL_UNIT_MASK | \
+ K7_EVNTSEL_EDGE_MASK | \
+ K7_EVNTSEL_INV_MASK | \
K7_EVNTSEL_COUNTER_MASK)
return event & K7_EVNTSEL_MASK;
* Can only be executed on the CPU where the counter is active.
* Returns the delta events processed.
*/
-static void
+static u64
x86_perf_counter_update(struct perf_counter *counter,
struct hw_perf_counter *hwc, int idx)
{
- u64 prev_raw_count, new_raw_count, delta;
+ int shift = 64 - x86_pmu.counter_bits;
+ u64 prev_raw_count, new_raw_count;
+ s64 delta;
/*
* Careful: an NMI might modify the previous counter value.
* (counter-)time and add that to the generic counter.
*
* Careful, not all hw sign-extends above the physical width
- * of the count, so we do that by clipping the delta to 32 bits:
+ * of the count.
*/
- delta = (u64)(u32)((s32)new_raw_count - (s32)prev_raw_count);
+ delta = (new_raw_count << shift) - (prev_raw_count << shift);
+ delta >>= shift;
atomic64_add(delta, &counter->count);
atomic64_sub(delta, &hwc->period_left);
+
+ return new_raw_count;
+}
+
+static atomic_t active_counters;
+static DEFINE_MUTEX(pmc_reserve_mutex);
+
+static bool reserve_pmc_hardware(void)
+{
+ int i;
+
+ if (nmi_watchdog == NMI_LOCAL_APIC)
+ disable_lapic_nmi_watchdog();
+
+ for (i = 0; i < x86_pmu.num_counters; i++) {
+ if (!reserve_perfctr_nmi(x86_pmu.perfctr + i))
+ goto perfctr_fail;
+ }
+
+ for (i = 0; i < x86_pmu.num_counters; i++) {
+ if (!reserve_evntsel_nmi(x86_pmu.eventsel + i))
+ goto eventsel_fail;
+ }
+
+ return true;
+
+eventsel_fail:
+ for (i--; i >= 0; i--)
+ release_evntsel_nmi(x86_pmu.eventsel + i);
+
+ i = x86_pmu.num_counters;
+
+perfctr_fail:
+ for (i--; i >= 0; i--)
+ release_perfctr_nmi(x86_pmu.perfctr + i);
+
+ if (nmi_watchdog == NMI_LOCAL_APIC)
+ enable_lapic_nmi_watchdog();
+
+ return false;
+}
+
+static void release_pmc_hardware(void)
+{
+ int i;
+
+ for (i = 0; i < x86_pmu.num_counters; i++) {
+ release_perfctr_nmi(x86_pmu.perfctr + i);
+ release_evntsel_nmi(x86_pmu.eventsel + i);
+ }
+
+ if (nmi_watchdog == NMI_LOCAL_APIC)
+ enable_lapic_nmi_watchdog();
+}
+
+static void hw_perf_counter_destroy(struct perf_counter *counter)
+{
+ if (atomic_dec_and_mutex_lock(&active_counters, &pmc_reserve_mutex)) {
+ release_pmc_hardware();
+ mutex_unlock(&pmc_reserve_mutex);
+ }
+}
+
+static inline int x86_pmu_initialized(void)
+{
+ return x86_pmu.handle_irq != NULL;
+}
+
+static inline int
+set_ext_hw_attr(struct hw_perf_counter *hwc, struct perf_counter_attr *attr)
+{
+ unsigned int cache_type, cache_op, cache_result;
+ u64 config, val;
+
+ config = attr->config;
+
+ cache_type = (config >> 0) & 0xff;
+ if (cache_type >= PERF_COUNT_HW_CACHE_MAX)
+ return -EINVAL;
+
+ cache_op = (config >> 8) & 0xff;
+ if (cache_op >= PERF_COUNT_HW_CACHE_OP_MAX)
+ return -EINVAL;
+
+ cache_result = (config >> 16) & 0xff;
+ if (cache_result >= PERF_COUNT_HW_CACHE_RESULT_MAX)
+ return -EINVAL;
+
+ val = hw_cache_event_ids[cache_type][cache_op][cache_result];
+
+ if (val == 0)
+ return -ENOENT;
+
+ if (val == -1)
+ return -EINVAL;
+
+ hwc->config |= val;
+
+ return 0;
}
/*
- * Setup the hardware configuration for a given hw_event_type
+ * Setup the hardware configuration for a given attr_type
*/
static int __hw_perf_counter_init(struct perf_counter *counter)
{
- struct perf_counter_hw_event *hw_event = &counter->hw_event;
+ struct perf_counter_attr *attr = &counter->attr;
struct hw_perf_counter *hwc = &counter->hw;
+ int err;
- if (unlikely(!perf_counters_initialized))
- return -EINVAL;
+ if (!x86_pmu_initialized())
+ return -ENODEV;
+
+ err = 0;
+ if (!atomic_inc_not_zero(&active_counters)) {
+ mutex_lock(&pmc_reserve_mutex);
+ if (atomic_read(&active_counters) == 0 && !reserve_pmc_hardware())
+ err = -EBUSY;
+ else
+ atomic_inc(&active_counters);
+ mutex_unlock(&pmc_reserve_mutex);
+ }
+ if (err)
+ return err;
/*
* Generate PMC IRQs:
/*
* Count user and OS events unless requested not to.
*/
- if (!hw_event->exclude_user)
+ if (!attr->exclude_user)
hwc->config |= ARCH_PERFMON_EVENTSEL_USR;
- if (!hw_event->exclude_kernel)
+ if (!attr->exclude_kernel)
hwc->config |= ARCH_PERFMON_EVENTSEL_OS;
- /*
- * If privileged enough, allow NMI events:
- */
- hwc->nmi = 0;
- if (capable(CAP_SYS_ADMIN) && hw_event->nmi)
- hwc->nmi = 1;
+ if (!hwc->sample_period) {
+ hwc->sample_period = x86_pmu.max_period;
+ hwc->last_period = hwc->sample_period;
+ atomic64_set(&hwc->period_left, hwc->sample_period);
+ }
+
+ counter->destroy = hw_perf_counter_destroy;
- hwc->irq_period = hw_event->irq_period;
/*
- * Intel PMCs cannot be accessed sanely above 32 bit width,
- * so we install an artificial 1<<31 period regardless of
- * the generic counter period:
+ * Raw event type provide the config in the event structure
*/
- if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
- if ((s64)hwc->irq_period <= 0 || hwc->irq_period > 0x7FFFFFFF)
- hwc->irq_period = 0x7FFFFFFF;
+ if (attr->type == PERF_TYPE_RAW) {
+ hwc->config |= x86_pmu.raw_event(attr->config);
+ return 0;
+ }
- atomic64_set(&hwc->period_left, hwc->irq_period);
+ if (attr->type == PERF_TYPE_HW_CACHE)
+ return set_ext_hw_attr(hwc, attr);
+ if (attr->config >= x86_pmu.max_events)
+ return -EINVAL;
/*
- * Raw event type provide the config in the event structure
+ * The generic map:
*/
- if (hw_event->raw) {
- hwc->config |= pmc_ops->raw_event(hw_event->type);
- } else {
- if (hw_event->type >= pmc_ops->max_events)
- return -EINVAL;
- /*
- * The generic map:
- */
- hwc->config |= pmc_ops->event_map(hw_event->type);
- }
- counter->wakeup_pending = 0;
+ hwc->config |= x86_pmu.event_map(attr->config);
return 0;
}
-static u64 pmc_intel_save_disable_all(void)
+static void intel_pmu_disable_all(void)
{
- u64 ctrl;
-
- rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0);
-
- return ctrl;
}
-static u64 pmc_amd_save_disable_all(void)
+static void amd_pmu_disable_all(void)
{
struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
- int enabled, idx;
+ int idx;
+
+ if (!cpuc->enabled)
+ return;
- enabled = cpuc->enabled;
cpuc->enabled = 0;
+ /*
+ * ensure we write the disable before we start disabling the
+ * counters proper, so that amd_pmu_enable_counter() does the
+ * right thing.
+ */
barrier();
- for (idx = 0; idx < nr_counters_generic; idx++) {
+ for (idx = 0; idx < x86_pmu.num_counters; idx++) {
u64 val;
+ if (!test_bit(idx, cpuc->active_mask))
+ continue;
rdmsrl(MSR_K7_EVNTSEL0 + idx, val);
- if (val & ARCH_PERFMON_EVENTSEL0_ENABLE) {
- val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE;
- wrmsrl(MSR_K7_EVNTSEL0 + idx, val);
- }
+ if (!(val & ARCH_PERFMON_EVENTSEL0_ENABLE))
+ continue;
+ val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE;
+ wrmsrl(MSR_K7_EVNTSEL0 + idx, val);
}
-
- return enabled;
}
-u64 hw_perf_save_disable(void)
+void hw_perf_disable(void)
{
- if (unlikely(!perf_counters_initialized))
- return 0;
-
- return pmc_ops->save_disable_all();
+ if (!x86_pmu_initialized())
+ return;
+ return x86_pmu.disable_all();
}
-/*
- * Exported because of ACPI idle
- */
-EXPORT_SYMBOL_GPL(hw_perf_save_disable);
-static void pmc_intel_restore_all(u64 ctrl)
+static void intel_pmu_enable_all(void)
{
- wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
+ wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, x86_pmu.intel_ctrl);
}
-static void pmc_amd_restore_all(u64 ctrl)
+static void amd_pmu_enable_all(void)
{
struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
int idx;
- cpuc->enabled = ctrl;
- barrier();
- if (!ctrl)
+ if (cpuc->enabled)
return;
- for (idx = 0; idx < nr_counters_generic; idx++) {
- if (test_bit(idx, cpuc->active_mask)) {
- u64 val;
+ cpuc->enabled = 1;
+ barrier();
- rdmsrl(MSR_K7_EVNTSEL0 + idx, val);
- val |= ARCH_PERFMON_EVENTSEL0_ENABLE;
- wrmsrl(MSR_K7_EVNTSEL0 + idx, val);
- }
+ for (idx = 0; idx < x86_pmu.num_counters; idx++) {
+ u64 val;
+
+ if (!test_bit(idx, cpuc->active_mask))
+ continue;
+ rdmsrl(MSR_K7_EVNTSEL0 + idx, val);
+ if (val & ARCH_PERFMON_EVENTSEL0_ENABLE)
+ continue;
+ val |= ARCH_PERFMON_EVENTSEL0_ENABLE;
+ wrmsrl(MSR_K7_EVNTSEL0 + idx, val);
}
}
-void hw_perf_restore(u64 ctrl)
+void hw_perf_enable(void)
{
- if (unlikely(!perf_counters_initialized))
+ if (!x86_pmu_initialized())
return;
-
- pmc_ops->restore_all(ctrl);
+ x86_pmu.enable_all();
}
-/*
- * Exported because of ACPI idle
- */
-EXPORT_SYMBOL_GPL(hw_perf_restore);
-static u64 pmc_intel_get_status(u64 mask)
+static inline u64 intel_pmu_get_status(void)
{
u64 status;
return status;
}
-static u64 pmc_amd_get_status(u64 mask)
-{
- u64 status = 0;
- int idx;
-
- for (idx = 0; idx < nr_counters_generic; idx++) {
- s64 val;
-
- if (!(mask & (1 << idx)))
- continue;
-
- rdmsrl(MSR_K7_PERFCTR0 + idx, val);
- val <<= (64 - counter_value_bits);
- if (val >= 0)
- status |= (1 << idx);
- }
-
- return status;
-}
-
-static u64 hw_perf_get_status(u64 mask)
-{
- if (unlikely(!perf_counters_initialized))
- return 0;
-
- return pmc_ops->get_status(mask);
-}
-
-static void pmc_intel_ack_status(u64 ack)
+static inline void intel_pmu_ack_status(u64 ack)
{
wrmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, ack);
}
-static void pmc_amd_ack_status(u64 ack)
-{
-}
-
-static void hw_perf_ack_status(u64 ack)
-{
- if (unlikely(!perf_counters_initialized))
- return;
-
- pmc_ops->ack_status(ack);
-}
-
-static void pmc_intel_enable(int idx, u64 config)
-{
- wrmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + idx,
- config | ARCH_PERFMON_EVENTSEL0_ENABLE);
-}
-
-static void pmc_amd_enable(int idx, u64 config)
-{
- struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
-
- set_bit(idx, cpuc->active_mask);
- if (cpuc->enabled)
- config |= ARCH_PERFMON_EVENTSEL0_ENABLE;
-
- wrmsrl(MSR_K7_EVNTSEL0 + idx, config);
-}
-
-static void hw_perf_enable(int idx, u64 config)
-{
- if (unlikely(!perf_counters_initialized))
- return;
-
- pmc_ops->enable(idx, config);
-}
-
-static void pmc_intel_disable(int idx, u64 config)
-{
- wrmsrl(MSR_ARCH_PERFMON_EVENTSEL0 + idx, config);
-}
-
-static void pmc_amd_disable(int idx, u64 config)
+static inline void x86_pmu_enable_counter(struct hw_perf_counter *hwc, int idx)
{
- struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
-
- clear_bit(idx, cpuc->active_mask);
- wrmsrl(MSR_K7_EVNTSEL0 + idx, config);
-
+ int err;
+ err = checking_wrmsrl(hwc->config_base + idx,
+ hwc->config | ARCH_PERFMON_EVENTSEL0_ENABLE);
}
-static void hw_perf_disable(int idx, u64 config)
+static inline void x86_pmu_disable_counter(struct hw_perf_counter *hwc, int idx)
{
- if (unlikely(!perf_counters_initialized))
- return;
-
- pmc_ops->disable(idx, config);
+ int err;
+ err = checking_wrmsrl(hwc->config_base + idx,
+ hwc->config);
}
static inline void
-__pmc_fixed_disable(struct perf_counter *counter,
- struct hw_perf_counter *hwc, unsigned int __idx)
+intel_pmu_disable_fixed(struct hw_perf_counter *hwc, int __idx)
{
int idx = __idx - X86_PMC_IDX_FIXED;
u64 ctrl_val, mask;
}
static inline void
-__pmc_generic_disable(struct perf_counter *counter,
- struct hw_perf_counter *hwc, unsigned int idx)
+intel_pmu_disable_counter(struct hw_perf_counter *hwc, int idx)
{
- if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL))
- __pmc_fixed_disable(counter, hwc, idx);
- else
- hw_perf_disable(idx, hwc->config);
+ if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) {
+ intel_pmu_disable_fixed(hwc, idx);
+ return;
+ }
+
+ x86_pmu_disable_counter(hwc, idx);
+}
+
+static inline void
+amd_pmu_disable_counter(struct hw_perf_counter *hwc, int idx)
+{
+ x86_pmu_disable_counter(hwc, idx);
}
static DEFINE_PER_CPU(u64, prev_left[X86_PMC_IDX_MAX]);
* Set the next IRQ period, based on the hwc->period_left value.
* To be called with the counter disabled in hw:
*/
-static void
-__hw_perf_counter_set_period(struct perf_counter *counter,
+static int
+x86_perf_counter_set_period(struct perf_counter *counter,
struct hw_perf_counter *hwc, int idx)
{
s64 left = atomic64_read(&hwc->period_left);
- s32 period = hwc->irq_period;
- int err;
+ s64 period = hwc->sample_period;
+ int err, ret = 0;
/*
* If we are way outside a reasoable range then just skip forward:
if (unlikely(left <= -period)) {
left = period;
atomic64_set(&hwc->period_left, left);
+ hwc->last_period = period;
+ ret = 1;
}
if (unlikely(left <= 0)) {
left += period;
atomic64_set(&hwc->period_left, left);
+ hwc->last_period = period;
+ ret = 1;
}
+ /*
+ * Quirk: certain CPUs dont like it if just 1 event is left:
+ */
+ if (unlikely(left < 2))
+ left = 2;
+
+ if (left > x86_pmu.max_period)
+ left = x86_pmu.max_period;
per_cpu(prev_left[idx], smp_processor_id()) = left;
atomic64_set(&hwc->prev_count, (u64)-left);
err = checking_wrmsrl(hwc->counter_base + idx,
- (u64)(-left) & counter_value_mask);
+ (u64)(-left) & x86_pmu.counter_mask);
+
+ return ret;
}
static inline void
-__pmc_fixed_enable(struct perf_counter *counter,
- struct hw_perf_counter *hwc, unsigned int __idx)
+intel_pmu_enable_fixed(struct hw_perf_counter *hwc, int __idx)
{
int idx = __idx - X86_PMC_IDX_FIXED;
u64 ctrl_val, bits, mask;
err = checking_wrmsrl(hwc->config_base, ctrl_val);
}
-static void
-__pmc_generic_enable(struct perf_counter *counter,
- struct hw_perf_counter *hwc, int idx)
+static void intel_pmu_enable_counter(struct hw_perf_counter *hwc, int idx)
{
- if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL))
- __pmc_fixed_enable(counter, hwc, idx);
+ if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) {
+ intel_pmu_enable_fixed(hwc, idx);
+ return;
+ }
+
+ x86_pmu_enable_counter(hwc, idx);
+}
+
+static void amd_pmu_enable_counter(struct hw_perf_counter *hwc, int idx)
+{
+ struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
+
+ if (cpuc->enabled)
+ x86_pmu_enable_counter(hwc, idx);
else
- hw_perf_enable(idx, hwc->config);
+ x86_pmu_disable_counter(hwc, idx);
}
static int
{
unsigned int event;
- if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
- return -1;
-
- if (unlikely(hwc->nmi))
+ if (!x86_pmu.num_counters_fixed)
return -1;
event = hwc->config & ARCH_PERFMON_EVENT_MASK;
- if (unlikely(event == pmc_ops->event_map(PERF_COUNT_INSTRUCTIONS)))
+ if (unlikely(event == x86_pmu.event_map(PERF_COUNT_HW_INSTRUCTIONS)))
return X86_PMC_IDX_FIXED_INSTRUCTIONS;
- if (unlikely(event == pmc_ops->event_map(PERF_COUNT_CPU_CYCLES)))
+ if (unlikely(event == x86_pmu.event_map(PERF_COUNT_HW_CPU_CYCLES)))
return X86_PMC_IDX_FIXED_CPU_CYCLES;
- if (unlikely(event == pmc_ops->event_map(PERF_COUNT_BUS_CYCLES)))
+ if (unlikely(event == x86_pmu.event_map(PERF_COUNT_HW_BUS_CYCLES)))
return X86_PMC_IDX_FIXED_BUS_CYCLES;
return -1;
/*
* Find a PMC slot for the freshly enabled / scheduled in counter:
*/
-static int pmc_generic_enable(struct perf_counter *counter)
+static int x86_pmu_enable(struct perf_counter *counter)
{
struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
struct hw_perf_counter *hwc = &counter->hw;
* Try to get the fixed counter, if that is already taken
* then try to get a generic counter:
*/
- if (test_and_set_bit(idx, cpuc->used))
+ if (test_and_set_bit(idx, cpuc->used_mask))
goto try_generic;
hwc->config_base = MSR_ARCH_PERFMON_FIXED_CTR_CTRL;
} else {
idx = hwc->idx;
/* Try to get the previous generic counter again */
- if (test_and_set_bit(idx, cpuc->used)) {
+ if (test_and_set_bit(idx, cpuc->used_mask)) {
try_generic:
- idx = find_first_zero_bit(cpuc->used, nr_counters_generic);
- if (idx == nr_counters_generic)
+ idx = find_first_zero_bit(cpuc->used_mask,
+ x86_pmu.num_counters);
+ if (idx == x86_pmu.num_counters)
return -EAGAIN;
- set_bit(idx, cpuc->used);
+ set_bit(idx, cpuc->used_mask);
hwc->idx = idx;
}
- hwc->config_base = pmc_ops->eventsel;
- hwc->counter_base = pmc_ops->perfctr;
+ hwc->config_base = x86_pmu.eventsel;
+ hwc->counter_base = x86_pmu.perfctr;
}
- perf_counters_lapic_init(hwc->nmi);
+ perf_counters_lapic_init();
- __pmc_generic_disable(counter, hwc, idx);
+ x86_pmu.disable(hwc, idx);
cpuc->counters[idx] = counter;
- /*
- * Make it visible before enabling the hw:
- */
- smp_wmb();
+ set_bit(idx, cpuc->active_mask);
- __hw_perf_counter_set_period(counter, hwc, idx);
- __pmc_generic_enable(counter, hwc, idx);
+ x86_perf_counter_set_period(counter, hwc, idx);
+ x86_pmu.enable(hwc, idx);
return 0;
}
+static void x86_pmu_unthrottle(struct perf_counter *counter)
+{
+ struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
+ struct hw_perf_counter *hwc = &counter->hw;
+
+ if (WARN_ON_ONCE(hwc->idx >= X86_PMC_IDX_MAX ||
+ cpuc->counters[hwc->idx] != counter))
+ return;
+
+ x86_pmu.enable(hwc, hwc->idx);
+}
+
void perf_counter_print_debug(void)
{
u64 ctrl, status, overflow, pmc_ctrl, pmc_count, prev_left, fixed;
struct cpu_hw_counters *cpuc;
+ unsigned long flags;
int cpu, idx;
- if (!nr_counters_generic)
+ if (!x86_pmu.num_counters)
return;
- local_irq_disable();
+ local_irq_save(flags);
cpu = smp_processor_id();
cpuc = &per_cpu(cpu_hw_counters, cpu);
- if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) {
+ if (x86_pmu.version >= 2) {
rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl);
rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status);
rdmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, overflow);
pr_info("CPU#%d: overflow: %016llx\n", cpu, overflow);
pr_info("CPU#%d: fixed: %016llx\n", cpu, fixed);
}
- pr_info("CPU#%d: used: %016llx\n", cpu, *(u64 *)cpuc->used);
+ pr_info("CPU#%d: used: %016llx\n", cpu, *(u64 *)cpuc->used_mask);
- for (idx = 0; idx < nr_counters_generic; idx++) {
- rdmsrl(pmc_ops->eventsel + idx, pmc_ctrl);
- rdmsrl(pmc_ops->perfctr + idx, pmc_count);
+ for (idx = 0; idx < x86_pmu.num_counters; idx++) {
+ rdmsrl(x86_pmu.eventsel + idx, pmc_ctrl);
+ rdmsrl(x86_pmu.perfctr + idx, pmc_count);
prev_left = per_cpu(prev_left[idx], cpu);
pr_info("CPU#%d: gen-PMC%d left: %016llx\n",
cpu, idx, prev_left);
}
- for (idx = 0; idx < nr_counters_fixed; idx++) {
+ for (idx = 0; idx < x86_pmu.num_counters_fixed; idx++) {
rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, pmc_count);
pr_info("CPU#%d: fixed-PMC%d count: %016llx\n",
cpu, idx, pmc_count);
}
- local_irq_enable();
+ local_irq_restore(flags);
}
-static void pmc_generic_disable(struct perf_counter *counter)
+static void x86_pmu_disable(struct perf_counter *counter)
{
struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters);
struct hw_perf_counter *hwc = &counter->hw;
- unsigned int idx = hwc->idx;
+ int idx = hwc->idx;
- __pmc_generic_disable(counter, hwc, idx);
+ /*
+ * Must be done before we disable, otherwise the nmi handler
+ * could reenable again:
+ */
+ clear_bit(idx, cpuc->active_mask);
+ x86_pmu.disable(hwc, idx);
- clear_bit(idx, cpuc->used);
- cpuc->counters[idx] = NULL;
/*
* Make sure the cleared pointer becomes visible before we
* (potentially) free the counter:
*/
- smp_wmb();
+ barrier();
/*
* Drain the remaining delta count out of a counter
* that we are disabling:
*/
x86_perf_counter_update(counter, hwc, idx);
-}
-
-static void perf_store_irq_data(struct perf_counter *counter, u64 data)
-{
- struct perf_data *irqdata = counter->irqdata;
-
- if (irqdata->len > PERF_DATA_BUFLEN - sizeof(u64)) {
- irqdata->overrun++;
- } else {
- u64 *p = (u64 *) &irqdata->data[irqdata->len];
-
- *p = data;
- irqdata->len += sizeof(u64);
- }
+ cpuc->counters[idx] = NULL;
+ clear_bit(idx, cpuc->used_mask);
}
/*
* Save and restart an expired counter. Called by NMI contexts,
* so it has to be careful about preempting normal counter ops:
*/
-static void perf_save_and_restart(struct perf_counter *counter)
+static int intel_pmu_save_and_restart(struct perf_counter *counter)
{
struct hw_perf_counter *hwc = &counter->hw;
int idx = hwc->idx;
+ int ret;
x86_perf_counter_update(counter, hwc, idx);
- __hw_perf_counter_set_period(counter, hwc, idx);
+ ret = x86_perf_counter_set_period(counter, hwc, idx);
if (counter->state == PERF_COUNTER_STATE_ACTIVE)
- __pmc_generic_enable(counter, hwc, idx);
+ intel_pmu_enable_counter(hwc, idx);
+
+ return ret;
}
-static void
-perf_handle_group(struct perf_counter *sibling, u64 *status, u64 *overflown)
+static void intel_pmu_reset(void)
{
- struct perf_counter *counter, *group_leader = sibling->group_leader;
+ unsigned long flags;
+ int idx;
- /*
- * Store sibling timestamps (if any):
- */
- list_for_each_entry(counter, &group_leader->sibling_list, list_entry) {
+ if (!x86_pmu.num_counters)
+ return;
- x86_perf_counter_update(counter, &counter->hw, counter->hw.idx);
- perf_store_irq_data(sibling, counter->hw_event.type);
- perf_store_irq_data(sibling, atomic64_read(&counter->count));
+ local_irq_save(flags);
+
+ printk("clearing PMU state on CPU#%d\n", smp_processor_id());
+
+ for (idx = 0; idx < x86_pmu.num_counters; idx++) {
+ checking_wrmsrl(x86_pmu.eventsel + idx, 0ull);
+ checking_wrmsrl(x86_pmu.perfctr + idx, 0ull);
+ }
+ for (idx = 0; idx < x86_pmu.num_counters_fixed; idx++) {
+ checking_wrmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, 0ull);
}
+
+ local_irq_restore(flags);
}
-/*
- * Maximum interrupt frequency of 100KHz per CPU
- */
-#define PERFMON_MAX_INTERRUPTS (100000/HZ)
/*
* This handler is triggered by the local APIC, so the APIC IRQ handling
* rules apply:
*/
-static int __smp_perf_counter_interrupt(struct pt_regs *regs, int nmi)
+static int intel_pmu_handle_irq(struct pt_regs *regs)
{
- int bit, cpu = smp_processor_id();
+ struct perf_sample_data data;
+ struct cpu_hw_counters *cpuc;
+ int bit, cpu, loops;
u64 ack, status;
- struct cpu_hw_counters *cpuc = &per_cpu(cpu_hw_counters, cpu);
- int ret = 0;
- cpuc->throttle_ctrl = hw_perf_save_disable();
+ data.regs = regs;
+ data.addr = 0;
- status = hw_perf_get_status(cpuc->throttle_ctrl);
- if (!status)
- goto out;
+ cpu = smp_processor_id();
+ cpuc = &per_cpu(cpu_hw_counters, cpu);
- ret = 1;
+ perf_disable();
+ status = intel_pmu_get_status();
+ if (!status) {
+ perf_enable();
+ return 0;
+ }
+
+ loops = 0;
again:
+ if (++loops > 100) {
+ WARN_ONCE(1, "perfcounters: irq loop stuck!\n");
+ perf_counter_print_debug();
+ intel_pmu_reset();
+ perf_enable();
+ return 1;
+ }
+
inc_irq_stat(apic_perf_irqs);
ack = status;
for_each_bit(bit, (unsigned long *)&status, X86_PMC_IDX_MAX) {
struct perf_counter *counter = cpuc->counters[bit];
clear_bit(bit, (unsigned long *) &status);
- if (!counter)
+ if (!test_bit(bit, cpuc->active_mask))
continue;
- perf_save_and_restart(counter);
-
- switch (counter->hw_event.record_type) {
- case PERF_RECORD_SIMPLE:
+ if (!intel_pmu_save_and_restart(counter))
continue;
- case PERF_RECORD_IRQ:
- perf_store_irq_data(counter, instruction_pointer(regs));
- break;
- case PERF_RECORD_GROUP:
- perf_handle_group(counter, &status, &ack);
- break;
- }
- /*
- * From NMI context we cannot call into the scheduler to
- * do a task wakeup - but we mark these generic as
- * wakeup_pending and initate a wakeup callback:
- */
- if (nmi) {
- counter->wakeup_pending = 1;
- set_tsk_thread_flag(current, TIF_PERF_COUNTERS);
- } else {
- wake_up(&counter->waitq);
- }
+
+ if (perf_counter_overflow(counter, 1, &data))
+ intel_pmu_disable_counter(&counter->hw, bit);
}
- hw_perf_ack_status(ack);
+ intel_pmu_ack_status(ack);
/*
* Repeat if there is more work to be done:
*/
- status = hw_perf_get_status(cpuc->throttle_ctrl);
+ status = intel_pmu_get_status();
if (status)
goto again;
-out:
- /*
- * Restore - do not reenable when global enable is off or throttled:
- */
- if (++cpuc->interrupts < PERFMON_MAX_INTERRUPTS)
- hw_perf_restore(cpuc->throttle_ctrl);
- return ret;
+ perf_enable();
+
+ return 1;
}
-void perf_counter_unthrottle(void)
+static int amd_pmu_handle_irq(struct pt_regs *regs)
{
+ struct perf_sample_data data;
struct cpu_hw_counters *cpuc;
+ struct perf_counter *counter;
+ struct hw_perf_counter *hwc;
+ int cpu, idx, handled = 0;
+ u64 val;
- if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
- return;
+ data.regs = regs;
+ data.addr = 0;
- if (unlikely(!perf_counters_initialized))
- return;
+ cpu = smp_processor_id();
+ cpuc = &per_cpu(cpu_hw_counters, cpu);
- cpuc = &__get_cpu_var(cpu_hw_counters);
- if (cpuc->interrupts >= PERFMON_MAX_INTERRUPTS) {
- if (printk_ratelimit())
- printk(KERN_WARNING "PERFMON: max interrupts exceeded!\n");
- hw_perf_restore(cpuc->throttle_ctrl);
+ for (idx = 0; idx < x86_pmu.num_counters; idx++) {
+ if (!test_bit(idx, cpuc->active_mask))
+ continue;
+
+ counter = cpuc->counters[idx];
+ hwc = &counter->hw;
+
+ val = x86_perf_counter_update(counter, hwc, idx);
+ if (val & (1ULL << (x86_pmu.counter_bits - 1)))
+ continue;
+
+ /*
+ * counter overflow
+ */
+ handled = 1;
+ data.period = counter->hw.last_period;
+
+ if (!x86_perf_counter_set_period(counter, hwc, idx))
+ continue;
+
+ if (perf_counter_overflow(counter, 1, &data))
+ amd_pmu_disable_counter(hwc, idx);
}
- cpuc->interrupts = 0;
+
+ if (handled)
+ inc_irq_stat(apic_perf_irqs);
+
+ return handled;
}
-void smp_perf_counter_interrupt(struct pt_regs *regs)
+void smp_perf_pending_interrupt(struct pt_regs *regs)
{
irq_enter();
- apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR);
ack_APIC_irq();
- __smp_perf_counter_interrupt(regs, 0);
+ inc_irq_stat(apic_pending_irqs);
+ perf_counter_do_pending();
irq_exit();
}
-/*
- * This handler is triggered by NMI contexts:
- */
-void perf_counter_notify(struct pt_regs *regs)
+void set_perf_counter_pending(void)
{
- struct cpu_hw_counters *cpuc;
- unsigned long flags;
- int bit, cpu;
-
- local_irq_save(flags);
- cpu = smp_processor_id();
- cpuc = &per_cpu(cpu_hw_counters, cpu);
-
- for_each_bit(bit, cpuc->used, X86_PMC_IDX_MAX) {
- struct perf_counter *counter = cpuc->counters[bit];
-
- if (!counter)
- continue;
-
- if (counter->wakeup_pending) {
- counter->wakeup_pending = 0;
- wake_up(&counter->waitq);
- }
- }
-
- local_irq_restore(flags);
+ apic->send_IPI_self(LOCAL_PENDING_VECTOR);
}
-void perf_counters_lapic_init(int nmi)
+void perf_counters_lapic_init(void)
{
- u32 apic_val;
-
- if (!perf_counters_initialized)
+ if (!x86_pmu_initialized())
return;
+
/*
- * Enable the performance counter vector in the APIC LVT:
+ * Always use NMI for PMU
*/
- apic_val = apic_read(APIC_LVTERR);
-
- apic_write(APIC_LVTERR, apic_val | APIC_LVT_MASKED);
- if (nmi)
- apic_write(APIC_LVTPC, APIC_DM_NMI);
- else
- apic_write(APIC_LVTPC, LOCAL_PERF_VECTOR);
- apic_write(APIC_LVTERR, apic_val);
+ apic_write(APIC_LVTPC, APIC_DM_NMI);
}
static int __kprobes
{
struct die_args *args = __args;
struct pt_regs *regs;
- int ret;
+
+ if (!atomic_read(&active_counters))
+ return NOTIFY_DONE;
switch (cmd) {
case DIE_NMI:
regs = args->regs;
apic_write(APIC_LVTPC, APIC_DM_NMI);
- ret = __smp_perf_counter_interrupt(regs, 1);
+ /*
+ * Can't rely on the handled return value to say it was our NMI, two
+ * counters could trigger 'simultaneously' raising two back-to-back NMIs.
+ *
+ * If the first NMI handles both, the latter will be empty and daze
+ * the CPU.
+ */
+ x86_pmu.handle_irq(regs);
- return ret ? NOTIFY_STOP : NOTIFY_OK;
+ return NOTIFY_STOP;
}
static __read_mostly struct notifier_block perf_counter_nmi_notifier = {
.priority = 1
};
-static struct pmc_x86_ops pmc_intel_ops = {
- .save_disable_all = pmc_intel_save_disable_all,
- .restore_all = pmc_intel_restore_all,
- .get_status = pmc_intel_get_status,
- .ack_status = pmc_intel_ack_status,
- .enable = pmc_intel_enable,
- .disable = pmc_intel_disable,
+static struct x86_pmu intel_pmu = {
+ .name = "Intel",
+ .handle_irq = intel_pmu_handle_irq,
+ .disable_all = intel_pmu_disable_all,
+ .enable_all = intel_pmu_enable_all,
+ .enable = intel_pmu_enable_counter,
+ .disable = intel_pmu_disable_counter,
.eventsel = MSR_ARCH_PERFMON_EVENTSEL0,
.perfctr = MSR_ARCH_PERFMON_PERFCTR0,
- .event_map = pmc_intel_event_map,
- .raw_event = pmc_intel_raw_event,
+ .event_map = intel_pmu_event_map,
+ .raw_event = intel_pmu_raw_event,
.max_events = ARRAY_SIZE(intel_perfmon_event_map),
+ /*
+ * Intel PMCs cannot be accessed sanely above 32 bit width,
+ * so we install an artificial 1<<31 period regardless of
+ * the generic counter period:
+ */
+ .max_period = (1ULL << 31) - 1,
};
-static struct pmc_x86_ops pmc_amd_ops = {
- .save_disable_all = pmc_amd_save_disable_all,
- .restore_all = pmc_amd_restore_all,
- .get_status = pmc_amd_get_status,
- .ack_status = pmc_amd_ack_status,
- .enable = pmc_amd_enable,
- .disable = pmc_amd_disable,
+static struct x86_pmu amd_pmu = {
+ .name = "AMD",
+ .handle_irq = amd_pmu_handle_irq,
+ .disable_all = amd_pmu_disable_all,
+ .enable_all = amd_pmu_enable_all,
+ .enable = amd_pmu_enable_counter,
+ .disable = amd_pmu_disable_counter,
.eventsel = MSR_K7_EVNTSEL0,
.perfctr = MSR_K7_PERFCTR0,
- .event_map = pmc_amd_event_map,
- .raw_event = pmc_amd_raw_event,
+ .event_map = amd_pmu_event_map,
+ .raw_event = amd_pmu_raw_event,
.max_events = ARRAY_SIZE(amd_perfmon_event_map),
+ .num_counters = 4,
+ .counter_bits = 48,
+ .counter_mask = (1ULL << 48) - 1,
+ /* use highest bit to detect overflow */
+ .max_period = (1ULL << 47) - 1,
};
-static struct pmc_x86_ops *pmc_intel_init(void)
+static int intel_pmu_init(void)
{
+ union cpuid10_edx edx;
union cpuid10_eax eax;
- unsigned int ebx;
unsigned int unused;
- union cpuid10_edx edx;
+ unsigned int ebx;
+ int version;
+
+ if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
+ return -ENODEV;
/*
* Check whether the Architectural PerfMon supports
*/
cpuid(10, &eax.full, &ebx, &unused, &edx.full);
if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED)
- return NULL;
+ return -ENODEV;
+
+ version = eax.split.version_id;
+ if (version < 2)
+ return -ENODEV;
+
+ x86_pmu = intel_pmu;
+ x86_pmu.version = version;
+ x86_pmu.num_counters = eax.split.num_counters;
+ x86_pmu.counter_bits = eax.split.bit_width;
+ x86_pmu.counter_mask = (1ULL << eax.split.bit_width) - 1;
- pr_info("Intel Performance Monitoring support detected.\n");
- pr_info("... version: %d\n", eax.split.version_id);
- pr_info("... bit width: %d\n", eax.split.bit_width);
- pr_info("... mask length: %d\n", eax.split.mask_length);
+ /*
+ * Quirk: v2 perfmon does not report fixed-purpose counters, so
+ * assume at least 3 counters:
+ */
+ x86_pmu.num_counters_fixed = max((int)edx.split.num_counters_fixed, 3);
+
+ rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, x86_pmu.intel_ctrl);
+
+ /*
+ * Install the hw-cache-events table:
+ */
+ switch (boot_cpu_data.x86_model) {
+ case 15: /* original 65 nm celeron/pentium/core2/xeon, "Merom"/"Conroe" */
+ case 22: /* single-core 65 nm celeron/core2solo "Merom-L"/"Conroe-L" */
+ case 23: /* current 45 nm celeron/core2/xeon "Penryn"/"Wolfdale" */
+ case 29: /* six-core 45 nm xeon "Dunnington" */
+ memcpy(hw_cache_event_ids, core2_hw_cache_event_ids,
+ sizeof(hw_cache_event_ids));
+
+ pr_cont("Core2 events, ");
+ break;
+ default:
+ case 26:
+ memcpy(hw_cache_event_ids, nehalem_hw_cache_event_ids,
+ sizeof(hw_cache_event_ids));
- nr_counters_generic = eax.split.num_counters;
- nr_counters_fixed = edx.split.num_counters_fixed;
- counter_value_mask = (1ULL << eax.split.bit_width) - 1;
+ pr_cont("Nehalem/Corei7 events, ");
+ break;
+ case 28:
+ memcpy(hw_cache_event_ids, atom_hw_cache_event_ids,
+ sizeof(hw_cache_event_ids));
- return &pmc_intel_ops;
+ pr_cont("Atom events, ");
+ break;
+ }
+ return 0;
}
-static struct pmc_x86_ops *pmc_amd_init(void)
+static int amd_pmu_init(void)
{
- nr_counters_generic = 4;
- nr_counters_fixed = 0;
- counter_value_mask = 0x0000FFFFFFFFFFFFULL;
- counter_value_bits = 48;
+ x86_pmu = amd_pmu;
- pr_info("AMD Performance Monitoring support detected.\n");
+ switch (boot_cpu_data.x86) {
+ case 0x0f:
+ case 0x10:
+ case 0x11:
+ memcpy(hw_cache_event_ids, amd_0f_hw_cache_event_ids,
+ sizeof(hw_cache_event_ids));
- return &pmc_amd_ops;
+ pr_cont("AMD Family 0f/10/11 events, ");
+ break;
+ }
+ return 0;
}
void __init init_hw_perf_counters(void)
{
- if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON))
- return;
+ int err;
+
+ pr_info("Performance Counters: ");
switch (boot_cpu_data.x86_vendor) {
case X86_VENDOR_INTEL:
- pmc_ops = pmc_intel_init();
+ err = intel_pmu_init();
break;
case X86_VENDOR_AMD:
- pmc_ops = pmc_amd_init();
+ err = amd_pmu_init();
break;
+ default:
+ return;
}
- if (!pmc_ops)
+ if (err != 0) {
+ pr_cont("no PMU driver, software counters only.\n");
return;
+ }
+
+ pr_cont("%s PMU driver.\n", x86_pmu.name);
- pr_info("... num counters: %d\n", nr_counters_generic);
- if (nr_counters_generic > X86_PMC_MAX_GENERIC) {
- nr_counters_generic = X86_PMC_MAX_GENERIC;
+ if (x86_pmu.num_counters > X86_PMC_MAX_GENERIC) {
+ x86_pmu.num_counters = X86_PMC_MAX_GENERIC;
WARN(1, KERN_ERR "hw perf counters %d > max(%d), clipping!",
- nr_counters_generic, X86_PMC_MAX_GENERIC);
+ x86_pmu.num_counters, X86_PMC_MAX_GENERIC);
}
- perf_counter_mask = (1 << nr_counters_generic) - 1;
- perf_max_counters = nr_counters_generic;
-
- pr_info("... value mask: %016Lx\n", counter_value_mask);
+ perf_counter_mask = (1 << x86_pmu.num_counters) - 1;
+ perf_max_counters = x86_pmu.num_counters;
- if (nr_counters_fixed > X86_PMC_MAX_FIXED) {
- nr_counters_fixed = X86_PMC_MAX_FIXED;
+ if (x86_pmu.num_counters_fixed > X86_PMC_MAX_FIXED) {
+ x86_pmu.num_counters_fixed = X86_PMC_MAX_FIXED;
WARN(1, KERN_ERR "hw perf counters fixed %d > max(%d), clipping!",
- nr_counters_fixed, X86_PMC_MAX_FIXED);
+ x86_pmu.num_counters_fixed, X86_PMC_MAX_FIXED);
}
- pr_info("... fixed counters: %d\n", nr_counters_fixed);
- perf_counter_mask |= ((1LL << nr_counters_fixed)-1) << X86_PMC_IDX_FIXED;
+ perf_counter_mask |=
+ ((1LL << x86_pmu.num_counters_fixed)-1) << X86_PMC_IDX_FIXED;
- pr_info("... counter mask: %016Lx\n", perf_counter_mask);
- perf_counters_initialized = true;
-
- perf_counters_lapic_init(0);
+ perf_counters_lapic_init();
register_die_notifier(&perf_counter_nmi_notifier);
+
+ pr_info("... version: %d\n", x86_pmu.version);
+ pr_info("... bit width: %d\n", x86_pmu.counter_bits);
+ pr_info("... generic counters: %d\n", x86_pmu.num_counters);
+ pr_info("... value mask: %016Lx\n", x86_pmu.counter_mask);
+ pr_info("... max period: %016Lx\n", x86_pmu.max_period);
+ pr_info("... fixed-purpose counters: %d\n", x86_pmu.num_counters_fixed);
+ pr_info("... counter mask: %016Lx\n", perf_counter_mask);
}
-static void pmc_generic_read(struct perf_counter *counter)
+static inline void x86_pmu_read(struct perf_counter *counter)
{
x86_perf_counter_update(counter, &counter->hw, counter->hw.idx);
}
-static const struct hw_perf_counter_ops x86_perf_counter_ops = {
- .enable = pmc_generic_enable,
- .disable = pmc_generic_disable,
- .read = pmc_generic_read,
+static const struct pmu pmu = {
+ .enable = x86_pmu_enable,
+ .disable = x86_pmu_disable,
+ .read = x86_pmu_read,
+ .unthrottle = x86_pmu_unthrottle,
};
-const struct hw_perf_counter_ops *
-hw_perf_counter_init(struct perf_counter *counter)
+const struct pmu *hw_perf_counter_init(struct perf_counter *counter)
{
int err;
err = __hw_perf_counter_init(counter);
if (err)
- return NULL;
+ return ERR_PTR(err);
+
+ return &pmu;
+}
+
+/*
+ * callchain support
+ */
+
+static inline
+void callchain_store(struct perf_callchain_entry *entry, unsigned long ip)
+{
+ if (entry->nr < MAX_STACK_DEPTH)
+ entry->ip[entry->nr++] = ip;
+}
+
+static DEFINE_PER_CPU(struct perf_callchain_entry, irq_entry);
+static DEFINE_PER_CPU(struct perf_callchain_entry, nmi_entry);
+
+
+static void
+backtrace_warning_symbol(void *data, char *msg, unsigned long symbol)
+{
+ /* Ignore warnings */
+}
+
+static void backtrace_warning(void *data, char *msg)
+{
+ /* Ignore warnings */
+}
+
+static int backtrace_stack(void *data, char *name)
+{
+ /* Don't bother with IRQ stacks for now */
+ return -1;
+}
+
+static void backtrace_address(void *data, unsigned long addr, int reliable)
+{
+ struct perf_callchain_entry *entry = data;
+
+ if (reliable)
+ callchain_store(entry, addr);
+}
+
+static const struct stacktrace_ops backtrace_ops = {
+ .warning = backtrace_warning,
+ .warning_symbol = backtrace_warning_symbol,
+ .stack = backtrace_stack,
+ .address = backtrace_address,
+};
+
+static void
+perf_callchain_kernel(struct pt_regs *regs, struct perf_callchain_entry *entry)
+{
+ unsigned long bp;
+ char *stack;
+ int nr = entry->nr;
+
+ callchain_store(entry, instruction_pointer(regs));
+
+ stack = ((char *)regs + sizeof(struct pt_regs));
+#ifdef CONFIG_FRAME_POINTER
+ bp = frame_pointer(regs);
+#else
+ bp = 0;
+#endif
+
+ dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, entry);
+
+ entry->kernel = entry->nr - nr;
+}
+
+
+struct stack_frame {
+ const void __user *next_fp;
+ unsigned long return_address;
+};
+
+static int copy_stack_frame(const void __user *fp, struct stack_frame *frame)
+{
+ int ret;
+
+ if (!access_ok(VERIFY_READ, fp, sizeof(*frame)))
+ return 0;
+
+ ret = 1;
+ pagefault_disable();
+ if (__copy_from_user_inatomic(frame, fp, sizeof(*frame)))
+ ret = 0;
+ pagefault_enable();
+
+ return ret;
+}
+
+static void
+perf_callchain_user(struct pt_regs *regs, struct perf_callchain_entry *entry)
+{
+ struct stack_frame frame;
+ const void __user *fp;
+ int nr = entry->nr;
+
+ regs = (struct pt_regs *)current->thread.sp0 - 1;
+ fp = (void __user *)regs->bp;
+
+ callchain_store(entry, regs->ip);
+
+ while (entry->nr < MAX_STACK_DEPTH) {
+ frame.next_fp = NULL;
+ frame.return_address = 0;
+
+ if (!copy_stack_frame(fp, &frame))
+ break;
+
+ if ((unsigned long)fp < user_stack_pointer(regs))
+ break;
+
+ callchain_store(entry, frame.return_address);
+ fp = frame.next_fp;
+ }
+
+ entry->user = entry->nr - nr;
+}
+
+static void
+perf_do_callchain(struct pt_regs *regs, struct perf_callchain_entry *entry)
+{
+ int is_user;
+
+ if (!regs)
+ return;
+
+ is_user = user_mode(regs);
+
+ if (!current || current->pid == 0)
+ return;
+
+ if (is_user && current->state != TASK_RUNNING)
+ return;
+
+ if (!is_user)
+ perf_callchain_kernel(regs, entry);
+
+ if (current->mm)
+ perf_callchain_user(regs, entry);
+}
+
+struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
+{
+ struct perf_callchain_entry *entry;
+
+ if (in_nmi())
+ entry = &__get_cpu_var(nmi_entry);
+ else
+ entry = &__get_cpu_var(irq_entry);
+
+ entry->nr = 0;
+ entry->hv = 0;
+ entry->kernel = 0;
+ entry->user = 0;
+
+ perf_do_callchain(regs, entry);
- return &x86_perf_counter_ops;
+ return entry;
}