struct debug_store *ds;
};
+struct event_constraint {
+ unsigned long idxmsk[BITS_TO_LONGS(X86_PMC_IDX_MAX)];
+ int code;
+};
+
+#define EVENT_CONSTRAINT(c, m) { .code = (c), .idxmsk[0] = (m) }
+#define EVENT_CONSTRAINT_END { .code = 0, .idxmsk[0] = 0 }
+
+#define for_each_event_constraint(e, c) \
+ for ((e) = (c); (e)->idxmsk[0]; (e)++)
+
+
/*
* struct x86_pmu - generic x86 pmu
*/
u64 intel_ctrl;
void (*enable_bts)(u64 config);
void (*disable_bts)(void);
+ int (*get_event_idx)(struct cpu_hw_events *cpuc,
+ struct hw_perf_event *hwc);
};
static struct x86_pmu x86_pmu __read_mostly;
.enabled = 1,
};
+static const struct event_constraint *event_constraints;
+
/*
* Not sure about some of these
*/
return hw_event & P6_EVNTSEL_MASK;
}
+static const struct event_constraint intel_p6_event_constraints[] =
+{
+ EVENT_CONSTRAINT(0xc1, 0x1), /* FLOPS */
+ EVENT_CONSTRAINT(0x10, 0x1), /* FP_COMP_OPS_EXE */
+ EVENT_CONSTRAINT(0x11, 0x1), /* FP_ASSIST */
+ EVENT_CONSTRAINT(0x12, 0x2), /* MUL */
+ EVENT_CONSTRAINT(0x13, 0x2), /* DIV */
+ EVENT_CONSTRAINT(0x14, 0x1), /* CYCLES_DIV_BUSY */
+ EVENT_CONSTRAINT_END
+};
/*
* Intel PerfMon v3. Used on Core2 and later.
[PERF_COUNT_HW_BUS_CYCLES] = 0x013c,
};
+static const struct event_constraint intel_core_event_constraints[] =
+{
+ EVENT_CONSTRAINT(0x10, 0x1), /* FP_COMP_OPS_EXE */
+ EVENT_CONSTRAINT(0x11, 0x2), /* FP_ASSIST */
+ EVENT_CONSTRAINT(0x12, 0x2), /* MUL */
+ EVENT_CONSTRAINT(0x13, 0x2), /* DIV */
+ EVENT_CONSTRAINT(0x14, 0x1), /* CYCLES_DIV_BUSY */
+ EVENT_CONSTRAINT(0x18, 0x1), /* IDLE_DURING_DIV */
+ EVENT_CONSTRAINT(0x19, 0x2), /* DELAYED_BYPASS */
+ EVENT_CONSTRAINT(0xa1, 0x1), /* RS_UOPS_DISPATCH_CYCLES */
+ EVENT_CONSTRAINT(0xcb, 0x1), /* MEM_LOAD_RETIRED */
+ EVENT_CONSTRAINT_END
+};
+
+static const struct event_constraint intel_nehalem_event_constraints[] =
+{
+ EVENT_CONSTRAINT(0x40, 0x3), /* L1D_CACHE_LD */
+ EVENT_CONSTRAINT(0x41, 0x3), /* L1D_CACHE_ST */
+ EVENT_CONSTRAINT(0x42, 0x3), /* L1D_CACHE_LOCK */
+ EVENT_CONSTRAINT(0x43, 0x3), /* L1D_ALL_REF */
+ EVENT_CONSTRAINT(0x4e, 0x3), /* L1D_PREFETCH */
+ EVENT_CONSTRAINT(0x4c, 0x3), /* LOAD_HIT_PRE */
+ EVENT_CONSTRAINT(0x51, 0x3), /* L1D */
+ EVENT_CONSTRAINT(0x52, 0x3), /* L1D_CACHE_PREFETCH_LOCK_FB_HIT */
+ EVENT_CONSTRAINT(0x53, 0x3), /* L1D_CACHE_LOCK_FB_HIT */
+ EVENT_CONSTRAINT(0xc5, 0x3), /* CACHE_LOCK_CYCLES */
+ EVENT_CONSTRAINT_END
+};
+
static u64 intel_pmu_event_map(int hw_event)
{
return intel_perfmon_event_map[hw_event];
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX];
-static const u64 nehalem_hw_cache_event_ids
+static __initconst u64 nehalem_hw_cache_event_ids
[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] =
},
};
-static const u64 core2_hw_cache_event_ids
+static __initconst u64 core2_hw_cache_event_ids
[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] =
},
};
-static const u64 atom_hw_cache_event_ids
+static __initconst u64 atom_hw_cache_event_ids
[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] =
#define CORE_EVNTSEL_UNIT_MASK 0x0000FF00ULL
#define CORE_EVNTSEL_EDGE_MASK 0x00040000ULL
#define CORE_EVNTSEL_INV_MASK 0x00800000ULL
-#define CORE_EVNTSEL_REG_MASK 0xFF000000ULL
+#define CORE_EVNTSEL_REG_MASK 0xFF000000ULL
#define CORE_EVNTSEL_MASK \
(CORE_EVNTSEL_EVENT_MASK | \
return hw_event & CORE_EVNTSEL_MASK;
}
-static const u64 amd_hw_cache_event_ids
+static __initconst u64 amd_hw_cache_event_ids
[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] =
*/
hwc->config = ARCH_PERFMON_EVENTSEL_INT;
+ hwc->idx = -1;
+
/*
* Count user and OS events unless requested not to.
*/
bits |= 0x2;
if (hwc->config & ARCH_PERFMON_EVENTSEL_OS)
bits |= 0x1;
+
+ /*
+ * ANY bit is supported in v3 and up
+ */
+ if (x86_pmu.version > 2 && hwc->config & ARCH_PERFMON_EVENTSEL_ANY)
+ bits |= 0x4;
+
bits <<= (idx * 4);
mask = 0xfULL << (idx * 4);
x86_pmu_enable_event(hwc, idx);
}
-static int
-fixed_mode_idx(struct perf_event *event, struct hw_perf_event *hwc)
+static int fixed_mode_idx(struct hw_perf_event *hwc)
{
unsigned int hw_event;
if (!x86_pmu.num_events_fixed)
return -1;
+ /*
+ * fixed counters do not take all possible filters
+ */
+ if (hwc->config & ARCH_PERFMON_EVENT_FILTER_MASK)
+ return -1;
+
if (unlikely(hw_event == x86_pmu.event_map(PERF_COUNT_HW_INSTRUCTIONS)))
return X86_PMC_IDX_FIXED_INSTRUCTIONS;
if (unlikely(hw_event == x86_pmu.event_map(PERF_COUNT_HW_CPU_CYCLES)))
}
/*
- * Find a PMC slot for the freshly enabled / scheduled in event:
+ * generic counter allocator: get next free counter
*/
-static int x86_pmu_enable(struct perf_event *event)
+static int
+gen_get_event_idx(struct cpu_hw_events *cpuc, struct hw_perf_event *hwc)
+{
+ int idx;
+
+ idx = find_first_zero_bit(cpuc->used_mask, x86_pmu.num_events);
+ return idx == x86_pmu.num_events ? -1 : idx;
+}
+
+/*
+ * intel-specific counter allocator: check event constraints
+ */
+static int
+intel_get_event_idx(struct cpu_hw_events *cpuc, struct hw_perf_event *hwc)
+{
+ const struct event_constraint *event_constraint;
+ int i, code;
+
+ if (!event_constraints)
+ goto skip;
+
+ code = hwc->config & CORE_EVNTSEL_EVENT_MASK;
+
+ for_each_event_constraint(event_constraint, event_constraints) {
+ if (code == event_constraint->code) {
+ for_each_bit(i, event_constraint->idxmsk, X86_PMC_IDX_MAX) {
+ if (!test_and_set_bit(i, cpuc->used_mask))
+ return i;
+ }
+ return -1;
+ }
+ }
+skip:
+ return gen_get_event_idx(cpuc, hwc);
+}
+
+static int
+x86_schedule_event(struct cpu_hw_events *cpuc, struct hw_perf_event *hwc)
{
- struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
- struct hw_perf_event *hwc = &event->hw;
int idx;
- idx = fixed_mode_idx(event, hwc);
+ idx = fixed_mode_idx(hwc);
if (idx == X86_PMC_IDX_FIXED_BTS) {
/* BTS is already occupied. */
if (test_and_set_bit(idx, cpuc->used_mask))
return -EAGAIN;
hwc->config_base = 0;
- hwc->event_base = 0;
+ hwc->event_base = 0;
hwc->idx = idx;
} else if (idx >= 0) {
/*
} else {
idx = hwc->idx;
/* Try to get the previous generic event again */
- if (test_and_set_bit(idx, cpuc->used_mask)) {
+ if (idx == -1 || test_and_set_bit(idx, cpuc->used_mask)) {
try_generic:
- idx = find_first_zero_bit(cpuc->used_mask,
- x86_pmu.num_events);
- if (idx == x86_pmu.num_events)
+ idx = x86_pmu.get_event_idx(cpuc, hwc);
+ if (idx == -1)
return -EAGAIN;
set_bit(idx, cpuc->used_mask);
hwc->idx = idx;
}
- hwc->config_base = x86_pmu.eventsel;
- hwc->event_base = x86_pmu.perfctr;
+ hwc->config_base = x86_pmu.eventsel;
+ hwc->event_base = x86_pmu.perfctr;
}
+ return idx;
+}
+
+/*
+ * Find a PMC slot for the freshly enabled / scheduled in event:
+ */
+static int x86_pmu_enable(struct perf_event *event)
+{
+ struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
+ struct hw_perf_event *hwc = &event->hw;
+ int idx;
+
+ idx = x86_schedule_event(cpuc, hwc);
+ if (idx < 0)
+ return idx;
+
perf_events_lapic_init();
x86_pmu.disable(hwc, idx);
data.period = event->hw.last_period;
data.addr = 0;
+ data.raw = NULL;
regs.ip = 0;
/*
u64 val;
data.addr = 0;
+ data.raw = NULL;
cpuc = &__get_cpu_var(cpu_hw_events);
u64 ack, status;
data.addr = 0;
+ data.raw = NULL;
cpuc = &__get_cpu_var(cpu_hw_events);
u64 val;
data.addr = 0;
+ data.raw = NULL;
cpuc = &__get_cpu_var(cpu_hw_events);
.priority = 1
};
-static struct x86_pmu p6_pmu = {
+static __initconst struct x86_pmu p6_pmu = {
.name = "p6",
.handle_irq = p6_pmu_handle_irq,
.disable_all = p6_pmu_disable_all,
*/
.event_bits = 32,
.event_mask = (1ULL << 32) - 1,
+ .get_event_idx = intel_get_event_idx,
};
-static struct x86_pmu intel_pmu = {
+static __initconst struct x86_pmu intel_pmu = {
.name = "Intel",
.handle_irq = intel_pmu_handle_irq,
.disable_all = intel_pmu_disable_all,
.max_period = (1ULL << 31) - 1,
.enable_bts = intel_pmu_enable_bts,
.disable_bts = intel_pmu_disable_bts,
+ .get_event_idx = intel_get_event_idx,
};
-static struct x86_pmu amd_pmu = {
+static __initconst struct x86_pmu amd_pmu = {
.name = "AMD",
.handle_irq = amd_pmu_handle_irq,
.disable_all = amd_pmu_disable_all,
.apic = 1,
/* use highest bit to detect overflow */
.max_period = (1ULL << 47) - 1,
+ .get_event_idx = gen_get_event_idx,
};
-static int p6_pmu_init(void)
+static __init int p6_pmu_init(void)
{
switch (boot_cpu_data.x86_model) {
case 1:
case 7:
case 8:
case 11: /* Pentium III */
+ event_constraints = intel_p6_event_constraints;
break;
case 9:
case 13:
/* Pentium M */
+ event_constraints = intel_p6_event_constraints;
break;
default:
pr_cont("unsupported p6 CPU model %d ",
x86_pmu = p6_pmu;
- if (!cpu_has_apic) {
- pr_info("no APIC, boot with the \"lapic\" boot parameter to force-enable it.\n");
- pr_info("no hardware sampling interrupt available.\n");
- x86_pmu.apic = 0;
- }
-
return 0;
}
-static int intel_pmu_init(void)
+static __init int intel_pmu_init(void)
{
union cpuid10_edx edx;
union cpuid10_eax eax;
sizeof(hw_cache_event_ids));
pr_cont("Core2 events, ");
+ event_constraints = intel_core_event_constraints;
break;
default:
case 26:
memcpy(hw_cache_event_ids, nehalem_hw_cache_event_ids,
sizeof(hw_cache_event_ids));
+ event_constraints = intel_nehalem_event_constraints;
pr_cont("Nehalem/Corei7 events, ");
break;
case 28:
return 0;
}
-static int amd_pmu_init(void)
+static __init int amd_pmu_init(void)
{
/* Performance-monitoring supported from K7 and later: */
if (boot_cpu_data.x86 < 6)
return 0;
}
+static void __init pmu_check_apic(void)
+{
+ if (cpu_has_apic)
+ return;
+
+ x86_pmu.apic = 0;
+ pr_info("no APIC, boot with the \"lapic\" boot parameter to force-enable it.\n");
+ pr_info("no hardware sampling interrupt available.\n");
+}
+
void __init init_hw_perf_events(void)
{
int err;
return;
}
+ pmu_check_apic();
+
pr_cont("%s PMU driver.\n", x86_pmu.name);
if (x86_pmu.num_events > X86_PMC_MAX_GENERIC) {
.unthrottle = x86_pmu_unthrottle,
};
+static int
+validate_event(struct cpu_hw_events *cpuc, struct perf_event *event)
+{
+ struct hw_perf_event fake_event = event->hw;
+
+ if (event->pmu && event->pmu != &pmu)
+ return 0;
+
+ return x86_schedule_event(cpuc, &fake_event) >= 0;
+}
+
+static int validate_group(struct perf_event *event)
+{
+ struct perf_event *sibling, *leader = event->group_leader;
+ struct cpu_hw_events fake_pmu;
+
+ memset(&fake_pmu, 0, sizeof(fake_pmu));
+
+ if (!validate_event(&fake_pmu, leader))
+ return -ENOSPC;
+
+ list_for_each_entry(sibling, &leader->sibling_list, group_entry) {
+ if (!validate_event(&fake_pmu, sibling))
+ return -ENOSPC;
+ }
+
+ if (!validate_event(&fake_pmu, event))
+ return -ENOSPC;
+
+ return 0;
+}
+
const struct pmu *hw_perf_event_init(struct perf_event *event)
{
int err;
err = __hw_perf_event_init(event);
+ if (!err) {
+ if (event->group_leader != event)
+ err = validate_group(event);
+ }
if (err) {
if (event->destroy)
event->destroy(event);
static DEFINE_PER_CPU(struct perf_callchain_entry, pmc_irq_entry);
static DEFINE_PER_CPU(struct perf_callchain_entry, pmc_nmi_entry);
-static DEFINE_PER_CPU(int, in_nmi_frame);
+static DEFINE_PER_CPU(int, in_ignored_frame);
static void
static int backtrace_stack(void *data, char *name)
{
- per_cpu(in_nmi_frame, smp_processor_id()) =
- x86_is_stack_id(NMI_STACK, name);
+ per_cpu(in_ignored_frame, smp_processor_id()) =
+ x86_is_stack_id(NMI_STACK, name) ||
+ x86_is_stack_id(DEBUG_STACK, name);
return 0;
}
{
struct perf_callchain_entry *entry = data;
- if (per_cpu(in_nmi_frame, smp_processor_id()))
+ if (per_cpu(in_ignored_frame, smp_processor_id()))
return;
if (reliable)
.warning_symbol = backtrace_warning_symbol,
.stack = backtrace_stack,
.address = backtrace_address,
+ .walk_stack = print_context_stack_bp,
};
#include "../dumpstack.h"
callchain_store(entry, PERF_CONTEXT_KERNEL);
callchain_store(entry, regs->ip);
- dump_trace(NULL, regs, NULL, 0, &backtrace_ops, entry);
+ dump_trace(NULL, regs, NULL, regs->bp, &backtrace_ops, entry);
}
/*
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff