static int perf_overcommit __read_mostly = 1;
static atomic_t nr_counters __read_mostly;
-static atomic_t nr_mmap_tracking __read_mostly;
-static atomic_t nr_munmap_tracking __read_mostly;
-static atomic_t nr_comm_tracking __read_mostly;
+static atomic_t nr_mmap_counters __read_mostly;
+static atomic_t nr_munmap_counters __read_mostly;
+static atomic_t nr_comm_counters __read_mostly;
int sysctl_perf_counter_priv __read_mostly; /* do we need to be privileged */
int sysctl_perf_counter_mlock __read_mostly = 512; /* 'free' kb per user */
int sysctl_perf_counter_limit __read_mostly = 100000; /* max NMIs per second */
+static atomic64_t perf_counter_id;
+
/*
* Lock for (sysadmin-configurable) counter reservations:
*/
if (!is_software_counter(counter))
cpuctx->active_oncpu--;
ctx->nr_active--;
- if (counter->hw_event.exclusive || !cpuctx->active_oncpu)
+ if (counter->attr.exclusive || !cpuctx->active_oncpu)
cpuctx->exclusive = 0;
}
list_for_each_entry(counter, &group_counter->sibling_list, list_entry)
counter_sched_out(counter, cpuctx, ctx);
- if (group_counter->hw_event.exclusive)
+ if (group_counter->attr.exclusive)
cpuctx->exclusive = 0;
}
cpuctx->active_oncpu++;
ctx->nr_active++;
- if (counter->hw_event.exclusive)
+ if (counter->attr.exclusive)
cpuctx->exclusive = 1;
return 0;
* If this group is exclusive and there are already
* counters on the CPU, it can't go on.
*/
- if (counter->hw_event.exclusive && cpuctx->active_oncpu)
+ if (counter->attr.exclusive && cpuctx->active_oncpu)
return 0;
/*
* Otherwise, try to add it if all previous groups were able
*/
if (leader != counter)
group_sched_out(leader, cpuctx, ctx);
- if (leader->hw_event.pinned) {
+ if (leader->attr.pinned) {
update_group_times(leader);
leader->state = PERF_COUNTER_STATE_ERROR;
}
*/
if (leader != counter)
group_sched_out(leader, cpuctx, ctx);
- if (leader->hw_event.pinned) {
+ if (leader->attr.pinned) {
update_group_times(leader);
leader->state = PERF_COUNTER_STATE_ERROR;
}
/*
* not supported on inherited counters
*/
- if (counter->hw_event.inherit)
+ if (counter->attr.inherit)
return -EINVAL;
atomic_add(refresh, &counter->event_limit);
*/
list_for_each_entry(counter, &ctx->counter_list, list_entry) {
if (counter->state <= PERF_COUNTER_STATE_OFF ||
- !counter->hw_event.pinned)
+ !counter->attr.pinned)
continue;
if (counter->cpu != -1 && counter->cpu != cpu)
continue;
* ignore pinned counters since we did them already.
*/
if (counter->state <= PERF_COUNTER_STATE_OFF ||
- counter->hw_event.pinned)
+ counter->attr.pinned)
continue;
/*
interrupts = 2*sysctl_perf_counter_limit/HZ;
}
- if (!counter->hw_event.freq || !counter->hw_event.sample_freq)
+ if (!counter->attr.freq || !counter->attr.sample_freq)
continue;
events = HZ * interrupts * counter->hw.sample_period;
- period = div64_u64(events, counter->hw_event.sample_freq);
+ period = div64_u64(events, counter->attr.sample_freq);
delta = (s64)(1 + period - counter->hw.sample_period);
delta >>= 1;
perf_pending_sync(counter);
atomic_dec(&nr_counters);
- if (counter->hw_event.mmap)
- atomic_dec(&nr_mmap_tracking);
- if (counter->hw_event.munmap)
- atomic_dec(&nr_munmap_tracking);
- if (counter->hw_event.comm)
- atomic_dec(&nr_comm_tracking);
+ if (counter->attr.mmap)
+ atomic_dec(&nr_mmap_counters);
+ if (counter->attr.munmap)
+ atomic_dec(&nr_munmap_counters);
+ if (counter->attr.comm)
+ atomic_dec(&nr_comm_counters);
if (counter->destroy)
counter->destroy(counter);
mutex_lock(&counter->child_mutex);
values[0] = perf_counter_read(counter);
n = 1;
- if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
+ if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
values[n++] = counter->total_time_enabled +
atomic64_read(&counter->child_total_time_enabled);
- if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
+ if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
values[n++] = counter->total_time_running +
atomic64_read(&counter->child_total_time_running);
- if (counter->hw_event.read_format & PERF_FORMAT_ID)
+ if (counter->attr.read_format & PERF_FORMAT_ID)
values[n++] = counter->id;
mutex_unlock(&counter->child_mutex);
mutex_unlock(&counter->child_mutex);
}
+static int perf_counter_period(struct perf_counter *counter, u64 __user *arg)
+{
+ struct perf_counter_context *ctx = counter->ctx;
+ unsigned long size;
+ int ret = 0;
+ u64 value;
+
+ if (!counter->attr.sample_period)
+ return -EINVAL;
+
+ size = copy_from_user(&value, arg, sizeof(value));
+ if (size != sizeof(value))
+ return -EFAULT;
+
+ if (!value)
+ return -EINVAL;
+
+ spin_lock_irq(&ctx->lock);
+ if (counter->attr.freq) {
+ if (value > sysctl_perf_counter_limit) {
+ ret = -EINVAL;
+ goto unlock;
+ }
+
+ counter->attr.sample_freq = value;
+ } else {
+ counter->attr.sample_period = value;
+ counter->hw.sample_period = value;
+
+ perf_log_period(counter, value);
+ }
+unlock:
+ spin_unlock_irq(&ctx->lock);
+
+ return ret;
+}
+
static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
{
struct perf_counter *counter = file->private_data;
case PERF_COUNTER_IOC_REFRESH:
return perf_counter_refresh(counter, arg);
+
+ case PERF_COUNTER_IOC_PERIOD:
+ return perf_counter_period(counter, (u64 __user *)arg);
+
default:
return -ENOTTY;
}
struct perf_counter *counter = handle->counter;
struct perf_mmap_data *data = handle->data;
- int wakeup_events = counter->hw_event.wakeup_events;
+ int wakeup_events = counter->attr.wakeup_events;
if (handle->overflow && wakeup_events) {
int events = atomic_inc_return(&data->events);
int nmi, struct pt_regs *regs, u64 addr)
{
int ret;
- u64 sample_type = counter->hw_event.sample_type;
+ u64 sample_type = counter->attr.sample_type;
struct perf_output_handle handle;
struct perf_event_header header;
u64 ip;
perf_output_put(&handle, addr);
if (sample_type & PERF_SAMPLE_CONFIG)
- perf_output_put(&handle, counter->hw_event.config);
+ perf_output_put(&handle, counter->attr.config);
if (sample_type & PERF_SAMPLE_CPU)
perf_output_put(&handle, cpu_entry);
perf_output_end(&handle);
}
+/*
+ * fork tracking
+ */
+
+struct perf_fork_event {
+ struct task_struct *task;
+
+ struct {
+ struct perf_event_header header;
+
+ u32 pid;
+ u32 ppid;
+ } event;
+};
+
+static void perf_counter_fork_output(struct perf_counter *counter,
+ struct perf_fork_event *fork_event)
+{
+ struct perf_output_handle handle;
+ int size = fork_event->event.header.size;
+ struct task_struct *task = fork_event->task;
+ int ret = perf_output_begin(&handle, counter, size, 0, 0);
+
+ if (ret)
+ return;
+
+ fork_event->event.pid = perf_counter_pid(counter, task);
+ fork_event->event.ppid = perf_counter_pid(counter, task->real_parent);
+
+ perf_output_put(&handle, fork_event->event);
+ perf_output_end(&handle);
+}
+
+static int perf_counter_fork_match(struct perf_counter *counter)
+{
+ if (counter->attr.comm || counter->attr.mmap || counter->attr.munmap)
+ return 1;
+
+ return 0;
+}
+
+static void perf_counter_fork_ctx(struct perf_counter_context *ctx,
+ struct perf_fork_event *fork_event)
+{
+ struct perf_counter *counter;
+
+ if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list))
+ return;
+
+ rcu_read_lock();
+ list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) {
+ if (perf_counter_fork_match(counter))
+ perf_counter_fork_output(counter, fork_event);
+ }
+ rcu_read_unlock();
+}
+
+static void perf_counter_fork_event(struct perf_fork_event *fork_event)
+{
+ struct perf_cpu_context *cpuctx;
+ struct perf_counter_context *ctx;
+
+ cpuctx = &get_cpu_var(perf_cpu_context);
+ perf_counter_fork_ctx(&cpuctx->ctx, fork_event);
+ put_cpu_var(perf_cpu_context);
+
+ rcu_read_lock();
+ /*
+ * doesn't really matter which of the child contexts the
+ * events ends up in.
+ */
+ ctx = rcu_dereference(current->perf_counter_ctxp);
+ if (ctx)
+ perf_counter_fork_ctx(ctx, fork_event);
+ rcu_read_unlock();
+}
+
+void perf_counter_fork(struct task_struct *task)
+{
+ struct perf_fork_event fork_event;
+
+ if (!atomic_read(&nr_comm_counters) &&
+ !atomic_read(&nr_mmap_counters) &&
+ !atomic_read(&nr_munmap_counters))
+ return;
+
+ fork_event = (struct perf_fork_event){
+ .task = task,
+ .event = {
+ .header = {
+ .type = PERF_EVENT_FORK,
+ .size = sizeof(fork_event.event),
+ },
+ },
+ };
+
+ perf_counter_fork_event(&fork_event);
+}
+
/*
* comm tracking
*/
perf_output_end(&handle);
}
-static int perf_counter_comm_match(struct perf_counter *counter,
- struct perf_comm_event *comm_event)
+static int perf_counter_comm_match(struct perf_counter *counter)
{
- if (counter->hw_event.comm &&
- comm_event->event.header.type == PERF_EVENT_COMM)
+ if (counter->attr.comm)
return 1;
return 0;
rcu_read_lock();
list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) {
- if (perf_counter_comm_match(counter, comm_event))
+ if (perf_counter_comm_match(counter))
perf_counter_comm_output(counter, comm_event);
}
rcu_read_unlock();
{
struct perf_comm_event comm_event;
- if (!atomic_read(&nr_comm_tracking))
+ if (!atomic_read(&nr_comm_counters))
return;
comm_event = (struct perf_comm_event){
static int perf_counter_mmap_match(struct perf_counter *counter,
struct perf_mmap_event *mmap_event)
{
- if (counter->hw_event.mmap &&
+ if (counter->attr.mmap &&
mmap_event->event.header.type == PERF_EVENT_MMAP)
return 1;
- if (counter->hw_event.munmap &&
+ if (counter->attr.munmap &&
mmap_event->event.header.type == PERF_EVENT_MUNMAP)
return 1;
{
struct perf_mmap_event mmap_event;
- if (!atomic_read(&nr_mmap_tracking))
+ if (!atomic_read(&nr_mmap_counters))
return;
mmap_event = (struct perf_mmap_event){
{
struct perf_mmap_event mmap_event;
- if (!atomic_read(&nr_munmap_tracking))
+ if (!atomic_read(&nr_munmap_counters))
return;
mmap_event = (struct perf_mmap_event){
if (!throttle) {
counter->hw.interrupts++;
- } else if (counter->hw.interrupts != MAX_INTERRUPTS) {
- counter->hw.interrupts++;
- if (HZ*counter->hw.interrupts > (u64)sysctl_perf_counter_limit) {
- counter->hw.interrupts = MAX_INTERRUPTS;
- perf_log_throttle(counter, 0);
+ } else {
+ if (counter->hw.interrupts != MAX_INTERRUPTS) {
+ counter->hw.interrupts++;
+ if (HZ*counter->hw.interrupts > (u64)sysctl_perf_counter_limit) {
+ counter->hw.interrupts = MAX_INTERRUPTS;
+ perf_log_throttle(counter, 0);
+ ret = 1;
+ }
+ } else {
+ /*
+ * Keep re-disabling counters even though on the previous
+ * pass we disabled it - just in case we raced with a
+ * sched-in and the counter got enabled again:
+ */
ret = 1;
}
}
* In case we exclude kernel IPs or are somehow not in interrupt
* context, provide the next best thing, the user IP.
*/
- if ((counter->hw_event.exclude_kernel || !regs) &&
- !counter->hw_event.exclude_user)
+ if ((counter->attr.exclude_kernel || !regs) &&
+ !counter->attr.exclude_user)
regs = task_pt_regs(current);
if (regs) {
if (!perf_swcounter_is_counting(counter))
return 0;
- if (counter->hw_event.config != event_config)
+ if (counter->attr.config != event_config)
return 0;
if (regs) {
- if (counter->hw_event.exclude_user && user_mode(regs))
+ if (counter->attr.exclude_user && user_mode(regs))
return 0;
- if (counter->hw_event.exclude_kernel && !user_mode(regs))
+ if (counter->attr.exclude_kernel && !user_mode(regs))
return 0;
}
static void tp_perf_counter_destroy(struct perf_counter *counter)
{
- ftrace_profile_disable(perf_event_id(&counter->hw_event));
+ ftrace_profile_disable(perf_event_id(&counter->attr));
}
static const struct pmu *tp_perf_counter_init(struct perf_counter *counter)
{
- int event_id = perf_event_id(&counter->hw_event);
+ int event_id = perf_event_id(&counter->attr);
int ret;
ret = ftrace_profile_enable(event_id);
return NULL;
counter->destroy = tp_perf_counter_destroy;
- counter->hw.sample_period = counter->hw_event.sample_period;
+ counter->hw.sample_period = counter->attr.sample_period;
return &perf_ops_generic;
}
* to be kernel events, and page faults are never hypervisor
* events.
*/
- switch (perf_event_id(&counter->hw_event)) {
+ switch (perf_event_id(&counter->attr)) {
case PERF_COUNT_CPU_CLOCK:
pmu = &perf_ops_cpu_clock;
* Allocate and initialize a counter structure
*/
static struct perf_counter *
-perf_counter_alloc(struct perf_counter_hw_event *hw_event,
+perf_counter_alloc(struct perf_counter_attr *attr,
int cpu,
struct perf_counter_context *ctx,
struct perf_counter *group_leader,
mutex_init(&counter->mmap_mutex);
- counter->cpu = cpu;
- counter->hw_event = *hw_event;
- counter->group_leader = group_leader;
- counter->pmu = NULL;
- counter->ctx = ctx;
- counter->oncpu = -1;
+ counter->cpu = cpu;
+ counter->attr = *attr;
+ counter->group_leader = group_leader;
+ counter->pmu = NULL;
+ counter->ctx = ctx;
+ counter->oncpu = -1;
- counter->state = PERF_COUNTER_STATE_INACTIVE;
- if (hw_event->disabled)
+ counter->ns = get_pid_ns(current->nsproxy->pid_ns);
+ counter->id = atomic64_inc_return(&perf_counter_id);
+
+ counter->state = PERF_COUNTER_STATE_INACTIVE;
+
+ if (attr->disabled)
counter->state = PERF_COUNTER_STATE_OFF;
pmu = NULL;
hwc = &counter->hw;
- if (hw_event->freq && hw_event->sample_freq)
- hwc->sample_period = div64_u64(TICK_NSEC, hw_event->sample_freq);
+ if (attr->freq && attr->sample_freq)
+ hwc->sample_period = div64_u64(TICK_NSEC, attr->sample_freq);
else
- hwc->sample_period = hw_event->sample_period;
+ hwc->sample_period = attr->sample_period;
/*
* we currently do not support PERF_SAMPLE_GROUP on inherited counters
*/
- if (hw_event->inherit && (hw_event->sample_type & PERF_SAMPLE_GROUP))
+ if (attr->inherit && (attr->sample_type & PERF_SAMPLE_GROUP))
goto done;
- if (perf_event_raw(hw_event)) {
+ if (perf_event_raw(attr)) {
pmu = hw_perf_counter_init(counter);
goto done;
}
- switch (perf_event_type(hw_event)) {
+ switch (perf_event_type(attr)) {
case PERF_TYPE_HARDWARE:
pmu = hw_perf_counter_init(counter);
break;
err = PTR_ERR(pmu);
if (err) {
+ if (counter->ns)
+ put_pid_ns(counter->ns);
kfree(counter);
return ERR_PTR(err);
}
counter->pmu = pmu;
atomic_inc(&nr_counters);
- if (counter->hw_event.mmap)
- atomic_inc(&nr_mmap_tracking);
- if (counter->hw_event.munmap)
- atomic_inc(&nr_munmap_tracking);
- if (counter->hw_event.comm)
- atomic_inc(&nr_comm_tracking);
+ if (counter->attr.mmap)
+ atomic_inc(&nr_mmap_counters);
+ if (counter->attr.munmap)
+ atomic_inc(&nr_munmap_counters);
+ if (counter->attr.comm)
+ atomic_inc(&nr_comm_counters);
return counter;
}
-static atomic64_t perf_counter_id;
-
/**
* sys_perf_counter_open - open a performance counter, associate it to a task/cpu
*
- * @hw_event_uptr: event type attributes for monitoring/sampling
+ * @attr_uptr: event type attributes for monitoring/sampling
* @pid: target pid
* @cpu: target cpu
* @group_fd: group leader counter fd
*/
SYSCALL_DEFINE5(perf_counter_open,
- const struct perf_counter_hw_event __user *, hw_event_uptr,
+ const struct perf_counter_attr __user *, attr_uptr,
pid_t, pid, int, cpu, int, group_fd, unsigned long, flags)
{
struct perf_counter *counter, *group_leader;
- struct perf_counter_hw_event hw_event;
+ struct perf_counter_attr attr;
struct perf_counter_context *ctx;
struct file *counter_file = NULL;
struct file *group_file = NULL;
if (flags)
return -EINVAL;
- if (copy_from_user(&hw_event, hw_event_uptr, sizeof(hw_event)) != 0)
+ if (copy_from_user(&attr, attr_uptr, sizeof(attr)) != 0)
return -EFAULT;
/*
/*
* Only a group leader can be exclusive or pinned
*/
- if (hw_event.exclusive || hw_event.pinned)
+ if (attr.exclusive || attr.pinned)
goto err_put_context;
}
- counter = perf_counter_alloc(&hw_event, cpu, ctx, group_leader,
+ counter = perf_counter_alloc(&attr, cpu, ctx, group_leader,
GFP_KERNEL);
ret = PTR_ERR(counter);
if (IS_ERR(counter))
list_add_tail(&counter->owner_entry, ¤t->perf_counter_list);
mutex_unlock(¤t->perf_counter_mutex);
- counter->ns = get_pid_ns(current->nsproxy->pid_ns);
- counter->id = atomic64_inc_return(&perf_counter_id);
-
fput_light(counter_file, fput_needed2);
out_fput:
if (parent_counter->parent)
parent_counter = parent_counter->parent;
- child_counter = perf_counter_alloc(&parent_counter->hw_event,
+ child_counter = perf_counter_alloc(&parent_counter->attr,
parent_counter->cpu, child_ctx,
group_leader, GFP_KERNEL);
if (IS_ERR(child_counter))
/*
* Make the child state follow the state of the parent counter,
- * not its hw_event.disabled bit. We hold the parent's mutex,
+ * not its attr.disabled bit. We hold the parent's mutex,
* so we won't race with perf_counter_{en, dis}able_family.
*/
if (parent_counter->state >= PERF_COUNTER_STATE_INACTIVE)
/*
* inherit into child's child as well:
*/
- child_counter->hw_event.inherit = 1;
+ child_counter->attr.inherit = 1;
/*
* Get a reference to the parent filp - we will fput it
if (counter != counter->group_leader)
continue;
- if (!counter->hw_event.inherit) {
+ if (!counter->attr.inherit) {
inherited_all = 0;
continue;
}
projects / karo-tx-linux.git / blobdiff