/* 0 == registered but off, 1 == registered and on */
static int nmi_enabled = 0;
+/* common functions */
+
+u64 op_x86_get_ctrl(struct op_x86_model_spec const *model,
+ struct op_counter_config *counter_config)
+{
+ u64 val = 0;
+ u16 event = (u16)counter_config->event;
+
+ val |= ARCH_PERFMON_EVENTSEL_INT;
+ val |= counter_config->user ? ARCH_PERFMON_EVENTSEL_USR : 0;
+ val |= counter_config->kernel ? ARCH_PERFMON_EVENTSEL_OS : 0;
+ val |= (counter_config->unit_mask & 0xFF) << 8;
+ event &= model->event_mask ? model->event_mask : 0xFF;
+ val |= event & 0xFF;
+ val |= (event & 0x0F00) << 24;
+
+ return val;
+}
+
+
static int profile_exceptions_notify(struct notifier_block *self,
unsigned long val, void *data)
{
switch (val) {
case DIE_NMI:
- if (model->check_ctrs(args->regs, &per_cpu(cpu_msrs, cpu)))
- ret = NOTIFY_STOP;
+ case DIE_NMI_IPI:
+ model->check_ctrs(args->regs, &per_cpu(cpu_msrs, cpu));
+ ret = NOTIFY_STOP;
break;
default:
break;
static void nmi_cpu_save_registers(struct op_msrs *msrs)
{
- unsigned int const nr_ctrs = model->num_counters;
- unsigned int const nr_ctrls = model->num_controls;
struct op_msr *counters = msrs->counters;
struct op_msr *controls = msrs->controls;
unsigned int i;
- for (i = 0; i < nr_ctrs; ++i) {
- if (counters[i].addr) {
- rdmsr(counters[i].addr,
- counters[i].saved.low,
- counters[i].saved.high);
- }
+ for (i = 0; i < model->num_counters; ++i) {
+ if (counters[i].addr)
+ rdmsrl(counters[i].addr, counters[i].saved);
}
- for (i = 0; i < nr_ctrls; ++i) {
- if (controls[i].addr) {
- rdmsr(controls[i].addr,
- controls[i].saved.low,
- controls[i].saved.high);
- }
+ for (i = 0; i < model->num_controls; ++i) {
+ if (controls[i].addr)
+ rdmsrl(controls[i].addr, controls[i].saved);
}
}
int cpu = smp_processor_id();
struct op_msrs *msrs = &per_cpu(cpu_msrs, cpu);
spin_lock(&oprofilefs_lock);
- model->setup_ctrs(msrs);
+ model->setup_ctrs(model, msrs);
spin_unlock(&oprofilefs_lock);
per_cpu(saved_lvtpc, cpu) = apic_read(APIC_LVTPC);
apic_write(APIC_LVTPC, APIC_DM_NMI);
static struct notifier_block profile_exceptions_nb = {
.notifier_call = profile_exceptions_notify,
.next = NULL,
- .priority = 0
+ .priority = 2
};
static int nmi_setup(void)
static void nmi_restore_registers(struct op_msrs *msrs)
{
- unsigned int const nr_ctrs = model->num_counters;
- unsigned int const nr_ctrls = model->num_controls;
struct op_msr *counters = msrs->counters;
struct op_msr *controls = msrs->controls;
unsigned int i;
- for (i = 0; i < nr_ctrls; ++i) {
- if (controls[i].addr) {
- wrmsr(controls[i].addr,
- controls[i].saved.low,
- controls[i].saved.high);
- }
+ for (i = 0; i < model->num_controls; ++i) {
+ if (controls[i].addr)
+ wrmsrl(controls[i].addr, controls[i].saved);
}
- for (i = 0; i < nr_ctrs; ++i) {
- if (counters[i].addr) {
- wrmsr(counters[i].addr,
- counters[i].saved.low,
- counters[i].saved.high);
- }
+ for (i = 0; i < model->num_counters; ++i) {
+ if (counters[i].addr)
+ wrmsrl(counters[i].addr, counters[i].saved);
}
}
*cpu_type = "i386/core_2";
break;
case 26:
- arch_perfmon_setup_counters();
+ model = &op_arch_perfmon_spec;
*cpu_type = "i386/core_i7";
break;
case 28:
return 1;
}
-static int __init arch_perfmon_init(char **cpu_type)
-{
- if (!cpu_has_arch_perfmon)
- return 0;
- *cpu_type = "i386/arch_perfmon";
- model = &op_arch_perfmon_spec;
- arch_perfmon_setup_counters();
- return 1;
-}
-
/* in order to get sysfs right */
static int using_nmi;
/* Needs to be at least an Athlon (or hammer in 32bit mode) */
switch (family) {
- default:
- return -ENODEV;
case 6:
- model = &op_amd_spec;
cpu_type = "i386/athlon";
break;
case 0xf:
- model = &op_amd_spec;
- /* Actually it could be i386/hammer too, but give
- user space an consistent name. */
+ /*
+ * Actually it could be i386/hammer too, but
+ * give user space an consistent name.
+ */
cpu_type = "x86-64/hammer";
break;
case 0x10:
- model = &op_amd_spec;
cpu_type = "x86-64/family10";
break;
case 0x11:
- model = &op_amd_spec;
cpu_type = "x86-64/family11h";
break;
+ default:
+ return -ENODEV;
}
+ model = &op_amd_spec;
break;
case X86_VENDOR_INTEL:
break;
}
- if (!cpu_type && !arch_perfmon_init(&cpu_type))
+ if (cpu_type)
+ break;
+
+ if (!cpu_has_arch_perfmon)
return -ENODEV;
+
+ /* use arch perfmon as fallback */
+ cpu_type = "i386/arch_perfmon";
+ model = &op_arch_perfmon_spec;
break;
default:
* @author Philippe Elie
* @author Graydon Hoare
* @author Andi Kleen
+ * @author Robert Richter <robert.richter@amd.com>
*/
#include <linux/oprofile.h>
#include <asm/msr.h>
#include <asm/apic.h>
#include <asm/nmi.h>
-#include <asm/perf_counter.h>
#include "op_x86_model.h"
#include "op_counter.h"
static int num_counters = 2;
static int counter_width = 32;
-#define CTR_IS_RESERVED(msrs, c) (msrs->counters[(c)].addr ? 1 : 0)
-#define CTR_OVERFLOWED(n) (!((n) & (1ULL<<(counter_width-1))))
-
-#define CTRL_IS_RESERVED(msrs, c) (msrs->controls[(c)].addr ? 1 : 0)
-#define CTRL_READ(l, h, msrs, c) do {rdmsr((msrs->controls[(c)].addr), (l), (h)); } while (0)
-#define CTRL_WRITE(l, h, msrs, c) do {wrmsr((msrs->controls[(c)].addr), (l), (h)); } while (0)
-#define CTRL_SET_ACTIVE(n) (n |= (1<<22))
-#define CTRL_SET_INACTIVE(n) (n &= ~(1<<22))
-#define CTRL_CLEAR(x) (x &= (1<<21))
-#define CTRL_SET_ENABLE(val) (val |= 1<<20)
-#define CTRL_SET_USR(val, u) (val |= ((u & 1) << 16))
-#define CTRL_SET_KERN(val, k) (val |= ((k & 1) << 17))
-#define CTRL_SET_UM(val, m) (val |= (m << 8))
-#define CTRL_SET_EVENT(val, e) (val |= e)
+#define MSR_PPRO_EVENTSEL_RESERVED ((0xFFFFFFFFULL<<32)|(1ULL<<21))
static u64 *reset_value;
}
-static void ppro_setup_ctrs(struct op_msrs const * const msrs)
+static void ppro_setup_ctrs(struct op_x86_model_spec const *model,
+ struct op_msrs const * const msrs)
{
- unsigned int low, high;
+ u64 val;
int i;
if (!reset_value) {
/* clear all counters */
for (i = 0 ; i < num_counters; ++i) {
- if (unlikely(!CTRL_IS_RESERVED(msrs, i)))
+ if (unlikely(!msrs->controls[i].addr))
continue;
- CTRL_READ(low, high, msrs, i);
- CTRL_CLEAR(low);
- CTRL_WRITE(low, high, msrs, i);
+ rdmsrl(msrs->controls[i].addr, val);
+ val &= model->reserved;
+ wrmsrl(msrs->controls[i].addr, val);
}
/* avoid a false detection of ctr overflows in NMI handler */
for (i = 0; i < num_counters; ++i) {
- if (unlikely(!CTR_IS_RESERVED(msrs, i)))
+ if (unlikely(!msrs->counters[i].addr))
continue;
wrmsrl(msrs->counters[i].addr, -1LL);
}
/* enable active counters */
for (i = 0; i < num_counters; ++i) {
- if ((counter_config[i].enabled) && (CTR_IS_RESERVED(msrs, i))) {
+ if (counter_config[i].enabled && msrs->counters[i].addr) {
reset_value[i] = counter_config[i].count;
-
wrmsrl(msrs->counters[i].addr, -reset_value[i]);
-
- CTRL_READ(low, high, msrs, i);
- CTRL_CLEAR(low);
- CTRL_SET_ENABLE(low);
- CTRL_SET_USR(low, counter_config[i].user);
- CTRL_SET_KERN(low, counter_config[i].kernel);
- CTRL_SET_UM(low, counter_config[i].unit_mask);
- CTRL_SET_EVENT(low, counter_config[i].event);
- CTRL_WRITE(low, high, msrs, i);
+ rdmsrl(msrs->controls[i].addr, val);
+ val &= model->reserved;
+ val |= op_x86_get_ctrl(model, &counter_config[i]);
+ wrmsrl(msrs->controls[i].addr, val);
} else {
reset_value[i] = 0;
}
u64 val;
int i;
+ /*
+ * This can happen if perf counters are in use when
+ * we steal the die notifier NMI.
+ */
+ if (unlikely(!reset_value))
+ goto out;
+
for (i = 0 ; i < num_counters; ++i) {
if (!reset_value[i])
continue;
rdmsrl(msrs->counters[i].addr, val);
- if (CTR_OVERFLOWED(val)) {
- oprofile_add_sample(regs, i);
- wrmsrl(msrs->counters[i].addr, -reset_value[i]);
- }
+ if (val & (1ULL << (counter_width - 1)))
+ continue;
+ oprofile_add_sample(regs, i);
+ wrmsrl(msrs->counters[i].addr, -reset_value[i]);
}
+ out:
/* Only P6 based Pentium M need to re-unmask the apic vector but it
* doesn't hurt other P6 variant */
apic_write(APIC_LVTPC, apic_read(APIC_LVTPC) & ~APIC_LVT_MASKED);
static void ppro_start(struct op_msrs const * const msrs)
{
- unsigned int low, high;
+ u64 val;
int i;
if (!reset_value)
return;
for (i = 0; i < num_counters; ++i) {
if (reset_value[i]) {
- CTRL_READ(low, high, msrs, i);
- CTRL_SET_ACTIVE(low);
- CTRL_WRITE(low, high, msrs, i);
+ rdmsrl(msrs->controls[i].addr, val);
+ val |= ARCH_PERFMON_EVENTSEL0_ENABLE;
+ wrmsrl(msrs->controls[i].addr, val);
}
}
}
static void ppro_stop(struct op_msrs const * const msrs)
{
- unsigned int low, high;
+ u64 val;
int i;
if (!reset_value)
for (i = 0; i < num_counters; ++i) {
if (!reset_value[i])
continue;
- CTRL_READ(low, high, msrs, i);
- CTRL_SET_INACTIVE(low);
- CTRL_WRITE(low, high, msrs, i);
+ rdmsrl(msrs->controls[i].addr, val);
+ val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE;
+ wrmsrl(msrs->controls[i].addr, val);
}
}
int i;
for (i = 0 ; i < num_counters ; ++i) {
- if (CTR_IS_RESERVED(msrs, i))
+ if (msrs->counters[i].addr)
release_perfctr_nmi(MSR_P6_PERFCTR0 + i);
}
for (i = 0 ; i < num_counters ; ++i) {
- if (CTRL_IS_RESERVED(msrs, i))
+ if (msrs->controls[i].addr)
release_evntsel_nmi(MSR_P6_EVNTSEL0 + i);
}
if (reset_value) {
}
-struct op_x86_model_spec op_ppro_spec = {
- .num_counters = 2, /* can be overriden */
- .num_controls = 2, /* dito */
+struct op_x86_model_spec const op_ppro_spec = {
+ .num_counters = 2,
+ .num_controls = 2,
+ .reserved = MSR_PPRO_EVENTSEL_RESERVED,
.fill_in_addresses = &ppro_fill_in_addresses,
.setup_ctrs = &ppro_setup_ctrs,
.check_ctrs = &ppro_check_ctrs,
* the specific CPU.
*/
-void arch_perfmon_setup_counters(void)
+static void arch_perfmon_setup_counters(void)
{
union cpuid10_eax eax;
op_arch_perfmon_spec.num_counters = num_counters;
op_arch_perfmon_spec.num_controls = num_counters;
- op_ppro_spec.num_counters = num_counters;
- op_ppro_spec.num_controls = num_counters;
+}
+
+static int arch_perfmon_init(struct oprofile_operations *ignore)
+{
+ arch_perfmon_setup_counters();
+ return 0;
}
struct op_x86_model_spec op_arch_perfmon_spec = {
+ .reserved = MSR_PPRO_EVENTSEL_RESERVED,
+ .init = &arch_perfmon_init,
/* num_counters/num_controls filled in at runtime */
.fill_in_addresses = &ppro_fill_in_addresses,
/* user space does the cpuid check for available events */
* @remark Read the file COPYING
*
* @author Graydon Hoare
+ * @author Robert Richter <robert.richter@amd.com>
*/
#ifndef OP_X86_MODEL_H
#define OP_X86_MODEL_H
-struct op_saved_msr {
- unsigned int high;
- unsigned int low;
-};
+#include <asm/types.h>
- #include <asm/intel_arch_perfmon.h>
++#include <asm/perf_counter.h>
struct op_msr {
- unsigned long addr;
- struct op_saved_msr saved;
+ unsigned long addr;
+ u64 saved;
};
struct op_msrs {
struct pt_regs;
+struct oprofile_operations;
+
/* The model vtable abstracts the differences between
* various x86 CPU models' perfctr support.
*/
struct op_x86_model_spec {
- int (*init)(struct oprofile_operations *ops);
- void (*exit)(void);
- unsigned int num_counters;
- unsigned int num_controls;
- void (*fill_in_addresses)(struct op_msrs * const msrs);
- void (*setup_ctrs)(struct op_msrs const * const msrs);
- int (*check_ctrs)(struct pt_regs * const regs,
- struct op_msrs const * const msrs);
- void (*start)(struct op_msrs const * const msrs);
- void (*stop)(struct op_msrs const * const msrs);
- void (*shutdown)(struct op_msrs const * const msrs);
+ unsigned int num_counters;
+ unsigned int num_controls;
+ u64 reserved;
+ u16 event_mask;
+ int (*init)(struct oprofile_operations *ops);
+ void (*exit)(void);
+ void (*fill_in_addresses)(struct op_msrs * const msrs);
+ void (*setup_ctrs)(struct op_x86_model_spec const *model,
+ struct op_msrs const * const msrs);
+ int (*check_ctrs)(struct pt_regs * const regs,
+ struct op_msrs const * const msrs);
+ void (*start)(struct op_msrs const * const msrs);
+ void (*stop)(struct op_msrs const * const msrs);
+ void (*shutdown)(struct op_msrs const * const msrs);
};
-extern struct op_x86_model_spec op_ppro_spec;
+struct op_counter_config;
+
+extern u64 op_x86_get_ctrl(struct op_x86_model_spec const *model,
+ struct op_counter_config *counter_config);
+
+extern struct op_x86_model_spec const op_ppro_spec;
extern struct op_x86_model_spec const op_p4_spec;
extern struct op_x86_model_spec const op_p4_ht2_spec;
extern struct op_x86_model_spec const op_amd_spec;
extern struct op_x86_model_spec op_arch_perfmon_spec;
-extern void arch_perfmon_setup_counters(void);
-
#endif /* OP_X86_MODEL_H */