1 files changed, 775 insertions, 313 deletions

diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index 1a933a221ea..534e20d14d6 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -42,6 +42,7 @@ static int perf_overcommit __read_mostly = 1;
 static atomic_t nr_counters __read_mostly;
 static atomic_t nr_mmap_counters __read_mostly;
 static atomic_t nr_comm_counters __read_mostly;
+static atomic_t nr_task_counters __read_mostly;
 
 /*
  * perf counter paranoia level:
@@ -87,6 +88,7 @@ void __weak hw_perf_disable(void)		{ barrier(); }
 void __weak hw_perf_enable(void)		{ barrier(); }
 
 void __weak hw_perf_counter_setup(int cpu)	{ barrier(); }
+void __weak hw_perf_counter_setup_online(int cpu)	{ barrier(); }
 
 int __weak
 hw_perf_group_sched_in(struct perf_counter *group_leader,
@@ -146,6 +148,28 @@ static void put_ctx(struct perf_counter_context *ctx)
 	}
 }
 
+static void unclone_ctx(struct perf_counter_context *ctx)
+{
+	if (ctx->parent_ctx) {
+		put_ctx(ctx->parent_ctx);
+		ctx->parent_ctx = NULL;
+	}
+}
+
+/*
+ * If we inherit counters we want to return the parent counter id
+ * to userspace.
+ */
+static u64 primary_counter_id(struct perf_counter *counter)
+{
+	u64 id = counter->id;
+
+	if (counter->parent)
+		id = counter->parent->id;
+
+	return id;
+}
+
 /*
  * Get the perf_counter_context for a task and lock it.
  * This has to cope with with the fact that until it is locked,
@@ -236,6 +260,8 @@ list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx)
 
 	list_add_rcu(&counter->event_entry, &ctx->event_list);
 	ctx->nr_counters++;
+	if (counter->attr.inherit_stat)
+		ctx->nr_stat++;
 }
 
 /*
@@ -250,6 +276,8 @@ list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx)
 	if (list_empty(&counter->list_entry))
 		return;
 	ctx->nr_counters--;
+	if (counter->attr.inherit_stat)
+		ctx->nr_stat--;
 
 	list_del_init(&counter->list_entry);
 	list_del_rcu(&counter->event_entry);
@@ -279,6 +307,10 @@ counter_sched_out(struct perf_counter *counter,
 		return;
 
 	counter->state = PERF_COUNTER_STATE_INACTIVE;
+	if (counter->pending_disable) {
+		counter->pending_disable = 0;
+		counter->state = PERF_COUNTER_STATE_OFF;
+	}
 	counter->tstamp_stopped = ctx->time;
 	counter->pmu->disable(counter);
 	counter->oncpu = -1;
@@ -1006,6 +1038,81 @@ static int context_equiv(struct perf_counter_context *ctx1,
 		&& !ctx1->pin_count && !ctx2->pin_count;
 }
 
+static void __perf_counter_read(void *counter);
+
+static void __perf_counter_sync_stat(struct perf_counter *counter,
+				     struct perf_counter *next_counter)
+{
+	u64 value;
+
+	if (!counter->attr.inherit_stat)
+		return;
+
+	/*
+	 * Update the counter value, we cannot use perf_counter_read()
+	 * because we're in the middle of a context switch and have IRQs
+	 * disabled, which upsets smp_call_function_single(), however
+	 * we know the counter must be on the current CPU, therefore we
+	 * don't need to use it.
+	 */
+	switch (counter->state) {
+	case PERF_COUNTER_STATE_ACTIVE:
+		__perf_counter_read(counter);
+		break;
+
+	case PERF_COUNTER_STATE_INACTIVE:
+		update_counter_times(counter);
+		break;
+
+	default:
+		break;
+	}
+
+	/*
+	 * In order to keep per-task stats reliable we need to flip the counter
+	 * values when we flip the contexts.
+	 */
+	value = atomic64_read(&next_counter->count);
+	value = atomic64_xchg(&counter->count, value);
+	atomic64_set(&next_counter->count, value);
+
+	swap(counter->total_time_enabled, next_counter->total_time_enabled);
+	swap(counter->total_time_running, next_counter->total_time_running);
+
+	/*
+	 * Since we swizzled the values, update the user visible data too.
+	 */
+	perf_counter_update_userpage(counter);
+	perf_counter_update_userpage(next_counter);
+}
+
+#define list_next_entry(pos, member) \
+	list_entry(pos->member.next, typeof(*pos), member)
+
+static void perf_counter_sync_stat(struct perf_counter_context *ctx,
+				   struct perf_counter_context *next_ctx)
+{
+	struct perf_counter *counter, *next_counter;
+
+	if (!ctx->nr_stat)
+		return;
+
+	counter = list_first_entry(&ctx->event_list,
+				   struct perf_counter, event_entry);
+
+	next_counter = list_first_entry(&next_ctx->event_list,
+					struct perf_counter, event_entry);
+
+	while (&counter->event_entry != &ctx->event_list &&
+	       &next_counter->event_entry != &next_ctx->event_list) {
+
+		__perf_counter_sync_stat(counter, next_counter);
+
+		counter = list_next_entry(counter, event_entry);
+		next_counter = list_next_entry(next_counter, event_entry);
+	}
+}
+
 /*
  * Called from scheduler to remove the counters of the current task,
  * with interrupts disabled.
@@ -1061,6 +1168,8 @@ void perf_counter_task_sched_out(struct task_struct *task,
 			ctx->task = next;
 			next_ctx->task = task;
 			do_switch = 0;
+
+			perf_counter_sync_stat(ctx, next_ctx);
 		}
 		spin_unlock(&next_ctx->lock);
 		spin_unlock(&ctx->lock);
@@ -1207,7 +1316,6 @@ static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu)
 #define MAX_INTERRUPTS (~0ULL)
 
 static void perf_log_throttle(struct perf_counter *counter, int enable);
-static void perf_log_period(struct perf_counter *counter, u64 period);
 
 static void perf_adjust_period(struct perf_counter *counter, u64 events)
 {
@@ -1226,8 +1334,6 @@ static void perf_adjust_period(struct perf_counter *counter, u64 events)
 	if (!sample_period)
 		sample_period = 1;
 
-	perf_log_period(counter, sample_period);
-
 	hwc->sample_period = sample_period;
 }
 
@@ -1348,9 +1454,54 @@ void perf_counter_task_tick(struct task_struct *curr, int cpu)
 }
 
 /*
+ * Enable all of a task's counters that have been marked enable-on-exec.
+ * This expects task == current.
+ */
+static void perf_counter_enable_on_exec(struct task_struct *task)
+{
+	struct perf_counter_context *ctx;
+	struct perf_counter *counter;
+	unsigned long flags;
+	int enabled = 0;
+
+	local_irq_save(flags);
+	ctx = task->perf_counter_ctxp;
+	if (!ctx || !ctx->nr_counters)
+		goto out;
+
+	__perf_counter_task_sched_out(ctx);
+
+	spin_lock(&ctx->lock);
+
+	list_for_each_entry(counter, &ctx->counter_list, list_entry) {
+		if (!counter->attr.enable_on_exec)
+			continue;
+		counter->attr.enable_on_exec = 0;
+		if (counter->state >= PERF_COUNTER_STATE_INACTIVE)
+			continue;
+		counter->state = PERF_COUNTER_STATE_INACTIVE;
+		counter->tstamp_enabled =
+			ctx->time - counter->total_time_enabled;
+		enabled = 1;
+	}
+
+	/*
+	 * Unclone this context if we enabled any counter.
+	 */
+	if (enabled)
+		unclone_ctx(ctx);
+
+	spin_unlock(&ctx->lock);
+
+	perf_counter_task_sched_in(task, smp_processor_id());
+ out:
+	local_irq_restore(flags);
+}
+
+/*
  * Cross CPU call to read the hardware counter
  */
-static void __read(void *info)
+static void __perf_counter_read(void *info)
 {
 	struct perf_counter *counter = info;
 	struct perf_counter_context *ctx = counter->ctx;
@@ -1372,7 +1523,7 @@ static u64 perf_counter_read(struct perf_counter *counter)
 	 */
 	if (counter->state == PERF_COUNTER_STATE_ACTIVE) {
 		smp_call_function_single(counter->oncpu,
-					 __read, counter, 1);
+					 __perf_counter_read, counter, 1);
 	} else if (counter->state == PERF_COUNTER_STATE_INACTIVE) {
 		update_counter_times(counter);
 	}
@@ -1398,7 +1549,6 @@ __perf_counter_init_context(struct perf_counter_context *ctx,
 
 static struct perf_counter_context *find_get_context(pid_t pid, int cpu)
 {
-	struct perf_counter_context *parent_ctx;
 	struct perf_counter_context *ctx;
 	struct perf_cpu_context *cpuctx;
 	struct task_struct *task;
@@ -1458,11 +1608,7 @@ static struct perf_counter_context *find_get_context(pid_t pid, int cpu)
  retry:
 	ctx = perf_lock_task_context(task, &flags);
 	if (ctx) {
-		parent_ctx = ctx->parent_ctx;
-		if (parent_ctx) {
-			put_ctx(parent_ctx);
-			ctx->parent_ctx = NULL;		/* no longer a clone */
-		}
+		unclone_ctx(ctx);
 		spin_unlock_irqrestore(&ctx->lock, flags);
 	}
 
@@ -1508,11 +1654,15 @@ static void free_counter(struct perf_counter *counter)
 {
 	perf_pending_sync(counter);
 
-	atomic_dec(&nr_counters);
-	if (counter->attr.mmap)
-		atomic_dec(&nr_mmap_counters);
-	if (counter->attr.comm)
-		atomic_dec(&nr_comm_counters);
+	if (!counter->parent) {
+		atomic_dec(&nr_counters);
+		if (counter->attr.mmap)
+			atomic_dec(&nr_mmap_counters);
+		if (counter->attr.comm)
+			atomic_dec(&nr_comm_counters);
+		if (counter->attr.task)
+			atomic_dec(&nr_task_counters);
+	}
 
 	if (counter->destroy)
 		counter->destroy(counter);
@@ -1546,14 +1696,133 @@ static int perf_release(struct inode *inode, struct file *file)
 	return 0;
 }
 
+static int perf_counter_read_size(struct perf_counter *counter)
+{
+	int entry = sizeof(u64); /* value */
+	int size = 0;
+	int nr = 1;
+
+	if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
+		size += sizeof(u64);
+
+	if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
+		size += sizeof(u64);
+
+	if (counter->attr.read_format & PERF_FORMAT_ID)
+		entry += sizeof(u64);
+
+	if (counter->attr.read_format & PERF_FORMAT_GROUP) {
+		nr += counter->group_leader->nr_siblings;
+		size += sizeof(u64);
+	}
+
+	size += entry * nr;
+
+	return size;
+}
+
+static u64 perf_counter_read_value(struct perf_counter *counter)
+{
+	struct perf_counter *child;
+	u64 total = 0;
+
+	total += perf_counter_read(counter);
+	list_for_each_entry(child, &counter->child_list, child_list)
+		total += perf_counter_read(child);
+
+	return total;
+}
+
+static int perf_counter_read_entry(struct perf_counter *counter,
+				   u64 read_format, char __user *buf)
+{
+	int n = 0, count = 0;
+	u64 values[2];
+
+	values[n++] = perf_counter_read_value(counter);
+	if (read_format & PERF_FORMAT_ID)
+		values[n++] = primary_counter_id(counter);
+
+	count = n * sizeof(u64);
+
+	if (copy_to_user(buf, values, count))
+		return -EFAULT;
+
+	return count;
+}
+
+static int perf_counter_read_group(struct perf_counter *counter,
+				   u64 read_format, char __user *buf)
+{
+	struct perf_counter *leader = counter->group_leader, *sub;
+	int n = 0, size = 0, err = -EFAULT;
+	u64 values[3];
+
+	values[n++] = 1 + leader->nr_siblings;
+	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) {
+		values[n++] = leader->total_time_enabled +
+			atomic64_read(&leader->child_total_time_enabled);
+	}
+	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) {
+		values[n++] = leader->total_time_running +
+			atomic64_read(&leader->child_total_time_running);
+	}
+
+	size = n * sizeof(u64);
+
+	if (copy_to_user(buf, values, size))
+		return -EFAULT;
+
+	err = perf_counter_read_entry(leader, read_format, buf + size);
+	if (err < 0)
+		return err;
+
+	size += err;
+
+	list_for_each_entry(sub, &leader->sibling_list, list_entry) {
+		err = perf_counter_read_entry(counter, read_format,
+				buf + size);
+		if (err < 0)
+			return err;
+
+		size += err;
+	}
+
+	return size;
+}
+
+static int perf_counter_read_one(struct perf_counter *counter,
+				 u64 read_format, char __user *buf)
+{
+	u64 values[4];
+	int n = 0;
+
+	values[n++] = perf_counter_read_value(counter);
+	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) {
+		values[n++] = counter->total_time_enabled +
+			atomic64_read(&counter->child_total_time_enabled);
+	}
+	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) {
+		values[n++] = counter->total_time_running +
+			atomic64_read(&counter->child_total_time_running);
+	}
+	if (read_format & PERF_FORMAT_ID)
+		values[n++] = primary_counter_id(counter);
+
+	if (copy_to_user(buf, values, n * sizeof(u64)))
+		return -EFAULT;
+
+	return n * sizeof(u64);
+}
+
 /*
  * Read the performance counter - simple non blocking version for now
  */
 static ssize_t
 perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count)
 {
-	u64 values[4];
-	int n;
+	u64 read_format = counter->attr.read_format;
+	int ret;
 
 	/*
 	 * Return end-of-file for a read on a counter that is in
@@ -1563,28 +1832,18 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count)
 	if (counter->state == PERF_COUNTER_STATE_ERROR)
 		return 0;
 
+	if (count < perf_counter_read_size(counter))
+		return -ENOSPC;
+
 	WARN_ON_ONCE(counter->ctx->parent_ctx);
 	mutex_lock(&counter->child_mutex);
-	values[0] = perf_counter_read(counter);
-	n = 1;
-	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->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
-		values[n++] = counter->total_time_running +
-			atomic64_read(&counter->child_total_time_running);
-	if (counter->attr.read_format & PERF_FORMAT_ID)
-		values[n++] = counter->id;
+	if (read_format & PERF_FORMAT_GROUP)
+		ret = perf_counter_read_group(counter, read_format, buf);
+	else
+		ret = perf_counter_read_one(counter, read_format, buf);
 	mutex_unlock(&counter->child_mutex);
 
-	if (count < n * sizeof(u64))
-		return -EINVAL;
-	count = n * sizeof(u64);
-
-	if (copy_to_user(buf, values, count))
-		return -EFAULT;
-
-	return count;
+	return ret;
 }
 
 static ssize_t
@@ -1681,8 +1940,6 @@ static int perf_counter_period(struct perf_counter *counter, u64 __user *arg)
 
 		counter->attr.sample_freq = value;
 	} else {
-		perf_log_period(counter, value);
-
 		counter->attr.sample_period = value;
 		counter->hw.sample_period = value;
 	}
@@ -1751,6 +2008,14 @@ int perf_counter_task_disable(void)
 	return 0;
 }
 
+static int perf_counter_index(struct perf_counter *counter)
+{
+	if (counter->state != PERF_COUNTER_STATE_ACTIVE)
+		return 0;
+
+	return counter->hw.idx + 1 - PERF_COUNTER_INDEX_OFFSET;
+}
+
 /*
  * Callers need to ensure there can be no nesting of this function, otherwise
  * the seqlock logic goes bad. We can not serialize this because the arch
@@ -1775,11 +2040,17 @@ void perf_counter_update_userpage(struct perf_counter *counter)
 	preempt_disable();
 	++userpg->lock;
 	barrier();
-	userpg->index = counter->hw.idx;
+	userpg->index = perf_counter_index(counter);
 	userpg->offset = atomic64_read(&counter->count);
 	if (counter->state == PERF_COUNTER_STATE_ACTIVE)
 		userpg->offset -= atomic64_read(&counter->hw.prev_count);
 
+	userpg->time_enabled = counter->total_time_enabled +
+			atomic64_read(&counter->child_total_time_enabled);
+
+	userpg->time_running = counter->total_time_running +
+			atomic64_read(&counter->child_total_time_running);
+
 	barrier();
 	++userpg->lock;
 	preempt_enable();
@@ -1876,7 +2147,7 @@ fail:
 
 static void perf_mmap_free_page(unsigned long addr)
 {
-	struct page *page = virt_to_page(addr);
+	struct page *page = virt_to_page((void *)addr);
 
 	page->mapping = NULL;
 	__free_page(page);
@@ -2076,7 +2347,7 @@ static void perf_pending_counter(struct perf_pending_entry *entry)
 
 	if (counter->pending_disable) {
 		counter->pending_disable = 0;
-		perf_counter_disable(counter);
+		__perf_counter_disable(counter);
 	}
 
 	if (counter->pending_wakeup) {
@@ -2461,7 +2732,80 @@ static u32 perf_counter_tid(struct perf_counter *counter, struct task_struct *p)
 	return task_pid_nr_ns(p, counter->ns);
 }
 
-static void perf_counter_output(struct perf_counter *counter, int nmi,
+static void perf_output_read_one(struct perf_output_handle *handle,
+				 struct perf_counter *counter)
+{
+	u64 read_format = counter->attr.read_format;
+	u64 values[4];
+	int n = 0;
+
+	values[n++] = atomic64_read(&counter->count);
+	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) {
+		values[n++] = counter->total_time_enabled +
+			atomic64_read(&counter->child_total_time_enabled);
+	}
+	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) {
+		values[n++] = counter->total_time_running +
+			atomic64_read(&counter->child_total_time_running);
+	}
+	if (read_format & PERF_FORMAT_ID)
+		values[n++] = primary_counter_id(counter);
+
+	perf_output_copy(handle, values, n * sizeof(u64));
+}
+
+/*
+ * XXX PERF_FORMAT_GROUP vs inherited counters seems difficult.
+ */
+static void perf_output_read_group(struct perf_output_handle *handle,
+			    struct perf_counter *counter)
+{
+	struct perf_counter *leader = counter->group_leader, *sub;
+	u64 read_format = counter->attr.read_format;
+	u64 values[5];
+	int n = 0;
+
+	values[n++] = 1 + leader->nr_siblings;
+
+	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
+		values[n++] = leader->total_time_enabled;
+
+	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
+		values[n++] = leader->total_time_running;
+
+	if (leader != counter)
+		leader->pmu->read(leader);
+
+	values[n++] = atomic64_read(&leader->count);
+	if (read_format & PERF_FORMAT_ID)
+		values[n++] = primary_counter_id(leader);
+
+	perf_output_copy(handle, values, n * sizeof(u64));
+
+	list_for_each_entry(sub, &leader->sibling_list, list_entry) {
+		n = 0;
+
+		if (sub != counter)
+			sub->pmu->read(sub);
+
+		values[n++] = atomic64_read(&sub->count);
+		if (read_format & PERF_FORMAT_ID)
+			values[n++] = primary_counter_id(sub);
+
+		perf_output_copy(handle, values, n * sizeof(u64));
+	}
+}
+
+static void perf_output_read(struct perf_output_handle *handle,
+			     struct perf_counter *counter)
+{
+	if (counter->attr.read_format & PERF_FORMAT_GROUP)
+		perf_output_read_group(handle, counter);
+	else
+		perf_output_read_one(handle, counter);
+}
+
+void perf_counter_output(struct perf_counter *counter, int nmi,
 				struct perf_sample_data *data)
 {
 	int ret;
@@ -2472,10 +2816,6 @@ static void perf_counter_output(struct perf_counter *counter, int nmi,
 	struct {
 		u32 pid, tid;
 	} tid_entry;
-	struct {
-		u64 id;
-		u64 counter;
-	} group_entry;
 	struct perf_callchain_entry *callchain = NULL;
 	int callchain_size = 0;
 	u64 time;
@@ -2483,15 +2823,14 @@ static void perf_counter_output(struct perf_counter *counter, int nmi,
 		u32 cpu, reserved;
 	} cpu_entry;
 
-	header.type = 0;
+	header.type = PERF_EVENT_SAMPLE;
 	header.size = sizeof(header);
 
-	header.misc = PERF_EVENT_MISC_OVERFLOW;
+	header.misc = 0;
 	header.misc |= perf_misc_flags(data->regs);
 
 	if (sample_type & PERF_SAMPLE_IP) {
 		ip = perf_instruction_pointer(data->regs);
-		header.type |= PERF_SAMPLE_IP;
 		header.size += sizeof(ip);
 	}
 
@@ -2500,7 +2839,6 @@ static void perf_counter_output(struct perf_counter *counter, int nmi,
 		tid_entry.pid = perf_counter_pid(counter, current);
 		tid_entry.tid = perf_counter_tid(counter, current);
 
-		header.type |= PERF_SAMPLE_TID;
 		header.size += sizeof(tid_entry);
 	}
 
@@ -2510,47 +2848,51 @@ static void perf_counter_output(struct perf_counter *counter, int nmi,
 		 */
 		time = sched_clock();
 
-		header.type |= PERF_SAMPLE_TIME;
 		header.size += sizeof(u64);
 	}
 
-	if (sample_type & PERF_SAMPLE_ADDR) {
-		header.type |= PERF_SAMPLE_ADDR;
+	if (sample_type & PERF_SAMPLE_ADDR)
 		header.size += sizeof(u64);
-	}
 
-	if (sample_type & PERF_SAMPLE_ID) {
-		header.type |= PERF_SAMPLE_ID;
+	if (sample_type & PERF_SAMPLE_ID)
+		header.size += sizeof(u64);
+
+	if (sample_type & PERF_SAMPLE_STREAM_ID)
 		header.size += sizeof(u64);
-	}
 
 	if (sample_type & PERF_SAMPLE_CPU) {
-		header.type |= PERF_SAMPLE_CPU;
 		header.size += sizeof(cpu_entry);
 
 		cpu_entry.cpu = raw_smp_processor_id();
+		cpu_entry.reserved = 0;
 	}
 
-	if (sample_type & PERF_SAMPLE_PERIOD) {
-		header.type |= PERF_SAMPLE_PERIOD;
+	if (sample_type & PERF_SAMPLE_PERIOD)
 		header.size += sizeof(u64);
-	}
 
-	if (sample_type & PERF_SAMPLE_GROUP) {
-		header.type |= PERF_SAMPLE_GROUP;
-		header.size += sizeof(u64) +
-			counter->nr_siblings * sizeof(group_entry);
-	}
+	if (sample_type & PERF_SAMPLE_READ)
+		header.size += perf_counter_read_size(counter);
 
 	if (sample_type & PERF_SAMPLE_CALLCHAIN) {
 		callchain = perf_callchain(data->regs);
 
 		if (callchain) {
 			callchain_size = (1 + callchain->nr) * sizeof(u64);
-
-			header.type |= PERF_SAMPLE_CALLCHAIN;
 			header.size += callchain_size;
-		}
+		} else
+			header.size += sizeof(u64);
+	}
+
+	if (sample_type & PERF_SAMPLE_RAW) {
+		int size = sizeof(u32);
+
+		if (data->raw)
+			size += data->raw->size;
+		else
+			size += sizeof(u32);
+
+		WARN_ON_ONCE(size & (sizeof(u64)-1));
+		header.size += size;
 	}
 
 	ret = perf_output_begin(&handle, counter, header.size, nmi, 1);
@@ -2571,7 +2913,13 @@ static void perf_counter_output(struct perf_counter *counter, int nmi,
 	if (sample_type & PERF_SAMPLE_ADDR)
 		perf_output_put(&handle, data->addr);
 
-	if (sample_type & PERF_SAMPLE_ID)
+	if (sample_type & PERF_SAMPLE_ID) {
+		u64 id = primary_counter_id(counter);
+
+		perf_output_put(&handle, id);
+	}
+
+	if (sample_type & PERF_SAMPLE_STREAM_ID)
 		perf_output_put(&handle, counter->id);
 
 	if (sample_type & PERF_SAMPLE_CPU)
@@ -2580,76 +2928,125 @@ static void perf_counter_output(struct perf_counter *counter, int nmi,
 	if (sample_type & PERF_SAMPLE_PERIOD)
 		perf_output_put(&handle, data->period);
 
-	/*
-	 * XXX PERF_SAMPLE_GROUP vs inherited counters seems difficult.
-	 */
-	if (sample_type & PERF_SAMPLE_GROUP) {
-		struct perf_counter *leader, *sub;
-		u64 nr = counter->nr_siblings;
+	if (sample_type & PERF_SAMPLE_READ)
+		perf_output_read(&handle, counter);
 
-		perf_output_put(&handle, nr);
-
-		leader = counter->group_leader;
-		list_for_each_entry(sub, &leader->sibling_list, list_entry) {
-			if (sub != counter)
-				sub->pmu->read(sub);
-
-			group_entry.id = sub->id;
-			group_entry.counter = atomic64_read(&sub->count);
+	if (sample_type & PERF_SAMPLE_CALLCHAIN) {
+		if (callchain)
+			perf_output_copy(&handle, callchain, callchain_size);
+		else {
+			u64 nr = 0;
+			perf_output_put(&handle, nr);
+		}
+	}
 
-			perf_output_put(&handle, group_entry);
+	if (sample_type & PERF_SAMPLE_RAW) {
+		if (data->raw) {
+			perf_output_put(&handle, data->raw->size);
+			perf_output_copy(&handle, data->raw->data, data->raw->size);
+		} else {
+			struct {
+				u32	size;
+				u32	data;
+			} raw = {
+				.size = sizeof(u32),
+				.data = 0,
+			};
+			perf_output_put(&handle, raw);
 		}
 	}
 
-	if (callchain)
-		perf_output_copy(&handle, callchain, callchain_size);
+	perf_output_end(&handle);
+}
+
+/*
+ * read event
+ */
+
+struct perf_read_event {
+	struct perf_event_header	header;
+
+	u32				pid;
+	u32				tid;
+};
+
+static void
+perf_counter_read_event(struct perf_counter *counter,
+			struct task_struct *task)
+{
+	struct perf_output_handle handle;
+	struct perf_read_event event = {
+		.header = {
+			.type = PERF_EVENT_READ,
+			.misc = 0,
+			.size = sizeof(event) + perf_counter_read_size(counter),
+		},
+		.pid = perf_counter_pid(counter, task),
+		.tid = perf_counter_tid(counter, task),
+	};
+	int ret;
+
+	ret = perf_output_begin(&handle, counter, event.header.size, 0, 0);
+	if (ret)
+		return;
+
+	perf_output_put(&handle, event);
+	perf_output_read(&handle, counter);
 
 	perf_output_end(&handle);
 }
 
 /*
- * fork tracking
+ * task tracking -- fork/exit
+ *
+ * enabled by: attr.comm | attr.mmap | attr.task
  */
 
-struct perf_fork_event {
-	struct task_struct	*task;
+struct perf_task_event {
+	struct task_struct		*task;
+	struct perf_counter_context	*task_ctx;
 
 	struct {
 		struct perf_event_header	header;
 
 		u32				pid;
 		u32				ppid;
+		u32				tid;
+		u32				ptid;
 	} event;
 };
 
-static void perf_counter_fork_output(struct perf_counter *counter,
-				     struct perf_fork_event *fork_event)
+static void perf_counter_task_output(struct perf_counter *counter,
+				     struct perf_task_event *task_event)
 {
 	struct perf_output_handle handle;
-	int size = fork_event->event.header.size;
-	struct task_struct *task = fork_event->task;
+	int size = task_event->event.header.size;
+	struct task_struct *task = task_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);
+	task_event->event.pid = perf_counter_pid(counter, task);
+	task_event->event.ppid = perf_counter_pid(counter, current);
+
+	task_event->event.tid = perf_counter_tid(counter, task);
+	task_event->event.ptid = perf_counter_tid(counter, current);
 
-	perf_output_put(&handle, fork_event->event);
+	perf_output_put(&handle, task_event->event);
 	perf_output_end(&handle);
 }
 
-static int perf_counter_fork_match(struct perf_counter *counter)
+static int perf_counter_task_match(struct perf_counter *counter)
 {
-	if (counter->attr.comm || counter->attr.mmap)
+	if (counter->attr.comm || counter->attr.mmap || counter->attr.task)
 		return 1;
 
 	return 0;
 }
 
-static void perf_counter_fork_ctx(struct perf_counter_context *ctx,
-				  struct perf_fork_event *fork_event)
+static void perf_counter_task_ctx(struct perf_counter_context *ctx,
+				  struct perf_task_event *task_event)
 {
 	struct perf_counter *counter;
 
@@ -2658,51 +3055,62 @@ static void perf_counter_fork_ctx(struct perf_counter_context *ctx,
 
 	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);
+		if (perf_counter_task_match(counter))
+			perf_counter_task_output(counter, task_event);
 	}
 	rcu_read_unlock();
 }
 
-static void perf_counter_fork_event(struct perf_fork_event *fork_event)
+static void perf_counter_task_event(struct perf_task_event *task_event)
 {
 	struct perf_cpu_context *cpuctx;
-	struct perf_counter_context *ctx;
+	struct perf_counter_context *ctx = task_event->task_ctx;
 
 	cpuctx = &get_cpu_var(perf_cpu_context);
-	perf_counter_fork_ctx(&cpuctx->ctx, fork_event);
+	perf_counter_task_ctx(&cpuctx->ctx, task_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)
+		ctx = rcu_dereference(task_event->task->perf_counter_ctxp);
 	if (ctx)
-		perf_counter_fork_ctx(ctx, fork_event);
+		perf_counter_task_ctx(ctx, task_event);
 	rcu_read_unlock();
 }
 
-void perf_counter_fork(struct task_struct *task)
+static void perf_counter_task(struct task_struct *task,
+			      struct perf_counter_context *task_ctx,
+			      int new)
 {
-	struct perf_fork_event fork_event;
+	struct perf_task_event task_event;
 
 	if (!atomic_read(&nr_comm_counters) &&
-	    !atomic_read(&nr_mmap_counters))
+	    !atomic_read(&nr_mmap_counters) &&
+	    !atomic_read(&nr_task_counters))
 		return;
 
-	fork_event = (struct perf_fork_event){
-		.task	= task,
-		.event  = {
+	task_event = (struct perf_task_event){
+		.task	  = task,
+		.task_ctx = task_ctx,
+		.event    = {
 			.header = {
-				.type = PERF_EVENT_FORK,
-				.size = sizeof(fork_event.event),
+				.type = new ? PERF_EVENT_FORK : PERF_EVENT_EXIT,
+				.misc = 0,
+				.size = sizeof(task_event.event),
 			},
+			/* .pid  */
+			/* .ppid */
+			/* .tid  */
+			/* .ptid */
 		},
 	};
 
-	perf_counter_fork_event(&fork_event);
+	perf_counter_task_event(&task_event);
+}
+
+void perf_counter_fork(struct task_struct *task)
+{
+	perf_counter_task(task, NULL, 1);
 }
 
 /*
@@ -2770,8 +3178,10 @@ static void perf_counter_comm_event(struct perf_comm_event *comm_event)
 	struct perf_cpu_context *cpuctx;
 	struct perf_counter_context *ctx;
 	unsigned int size;
-	char *comm = comm_event->task->comm;
+	char comm[TASK_COMM_LEN];
 
+	memset(comm, 0, sizeof(comm));
+	strncpy(comm, comm_event->task->comm, sizeof(comm));
 	size = ALIGN(strlen(comm)+1, sizeof(u64));
 
 	comm_event->comm = comm;
@@ -2798,13 +3208,24 @@ void perf_counter_comm(struct task_struct *task)
 {
 	struct perf_comm_event comm_event;
 
+	if (task->perf_counter_ctxp)
+		perf_counter_enable_on_exec(task);
+
 	if (!atomic_read(&nr_comm_counters))
 		return;
 
 	comm_event = (struct perf_comm_event){
 		.task	= task,
+		/* .comm      */
+		/* .comm_size */
 		.event  = {
-			.header = { .type = PERF_EVENT_COMM, },
+			.header = {
+				.type = PERF_EVENT_COMM,
+				.misc = 0,
+				/* .size */
+			},
+			/* .pid */
+			/* .tid */
 		},
 	};
 
@@ -2887,8 +3308,15 @@ static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event)
 	char *buf = NULL;
 	const char *name;
 
+	memset(tmp, 0, sizeof(tmp));
+
 	if (file) {
-		buf = kzalloc(PATH_MAX, GFP_KERNEL);
+		/*
+		 * d_path works from the end of the buffer backwards, so we
+		 * need to add enough zero bytes after the string to handle
+		 * the 64bit alignment we do later.
+		 */
+		buf = kzalloc(PATH_MAX + sizeof(u64), GFP_KERNEL);
 		if (!buf) {
 			name = strncpy(tmp, "//enomem", sizeof(tmp));
 			goto got_name;
@@ -2899,9 +3327,11 @@ static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event)
 			goto got_name;
 		}
 	} else {
-		name = arch_vma_name(mmap_event->vma);
-		if (name)
+		if (arch_vma_name(mmap_event->vma)) {
+			name = strncpy(tmp, arch_vma_name(mmap_event->vma),
+				       sizeof(tmp));
 			goto got_name;
+		}
 
 		if (!vma->vm_mm) {
 			name = strncpy(tmp, "[vdso]", sizeof(tmp));
@@ -2946,8 +3376,16 @@ void __perf_counter_mmap(struct vm_area_struct *vma)
 
 	mmap_event = (struct perf_mmap_event){
 		.vma	= vma,
+		/* .file_name */
+		/* .file_size */
 		.event  = {
-			.header = { .type = PERF_EVENT_MMAP, },
+			.header = {
+				.type = PERF_EVENT_MMAP,
+				.misc = 0,
+				/* .size */
+			},
+			/* .pid */
+			/* .tid */
 			.start  = vma->vm_start,
 			.len    = vma->vm_end - vma->vm_start,
 			.pgoff  = vma->vm_pgoff,
@@ -2958,49 +3396,6 @@ void __perf_counter_mmap(struct vm_area_struct *vma)
 }
 
 /*
- * Log sample_period changes so that analyzing tools can re-normalize the
- * event flow.
- */
-
-struct freq_event {
-	struct perf_event_header	header;
-	u64				time;
-	u64				id;
-	u64				period;
-};
-
-static void perf_log_period(struct perf_counter *counter, u64 period)
-{
-	struct perf_output_handle handle;
-	struct freq_event event;
-	int ret;
-
-	if (counter->hw.sample_period == period)
-		return;
-
-	if (counter->attr.sample_type & PERF_SAMPLE_PERIOD)
-		return;
-
-	event = (struct freq_event) {
-		.header = {
-			.type = PERF_EVENT_PERIOD,
-			.misc = 0,
-			.size = sizeof(event),
-		},
-		.time = sched_clock(),
-		.id = counter->id,
-		.period = period,
-	};
-
-	ret = perf_output_begin(&handle, counter, sizeof(event), 1, 0);
-	if (ret)
-		return;
-
-	perf_output_put(&handle, event);
-	perf_output_end(&handle);
-}
-
-/*
  * IRQ throttle logging
  */
 
@@ -3013,16 +3408,21 @@ static void perf_log_throttle(struct perf_counter *counter, int enable)
 		struct perf_event_header	header;
 		u64				time;
 		u64				id;
+		u64				stream_id;
 	} throttle_event = {
 		.header = {
-			.type = PERF_EVENT_THROTTLE + 1,
+			.type = PERF_EVENT_THROTTLE,
 			.misc = 0,
 			.size = sizeof(throttle_event),
 		},
-		.time	= sched_clock(),
-		.id	= counter->id,
+		.time		= sched_clock(),
+		.id		= primary_counter_id(counter),
+		.stream_id	= counter->id,
 	};
 
+	if (enable)
+		throttle_event.header.type = PERF_EVENT_UNTHROTTLE;
+
 	ret = perf_output_begin(&handle, counter, sizeof(throttle_event), 1, 0);
 	if (ret)
 		return;
@@ -3099,125 +3499,111 @@ int perf_counter_overflow(struct perf_counter *counter, int nmi,
  * Generic software counter infrastructure
  */
 
-static void perf_swcounter_update(struct perf_counter *counter)
+/*
+ * We directly increment counter->count and keep a second value in
+ * counter->hw.period_left to count intervals. This period counter
+ * is kept in the range [-sample_period, 0] so that we can use the
+ * sign as trigger.
+ */
+
+static u64 perf_swcounter_set_period(struct perf_counter *counter)
 {
 	struct hw_perf_counter *hwc = &counter->hw;
-	u64 prev, now;
-	s64 delta;
+	u64 period = hwc->last_period;
+	u64 nr, offset;
+	s64 old, val;
+
+	hwc->last_period = hwc->sample_period;
 
 again:
-	prev = atomic64_read(&hwc->prev_count);
-	now = atomic64_read(&hwc->count);
-	if (atomic64_cmpxchg(&hwc->prev_count, prev, now) != prev)
-		goto again;
+	old = val = atomic64_read(&hwc->period_left);
+	if (val < 0)
+		return 0;
 
-	delta = now - prev;
+	nr = div64_u64(period + val, period);
+	offset = nr * period;
+	val -= offset;
+	if (atomic64_cmpxchg(&hwc->period_left, old, val) != old)
+		goto again;
 
-	atomic64_add(delta, &counter->count);
-	atomic64_sub(delta, &hwc->period_left);
+	return nr;
 }
 
-static void perf_swcounter_set_period(struct perf_counter *counter)
+static void perf_swcounter_overflow(struct perf_counter *counter,
+				    int nmi, struct perf_sample_data *data)
 {
 	struct hw_perf_counter *hwc = &counter->hw;
-	s64 left = atomic64_read(&hwc->period_left);
-	s64 period = hwc->sample_period;
+	u64 overflow;
 
-	if (unlikely(left <= -period)) {
-		left = period;
-		atomic64_set(&hwc->period_left, left);
-		hwc->last_period = period;
-	}
+	data->period = counter->hw.last_