perf/core: Add a new read format to get a number of lost samples

Sometimes we want to know an accurate number of samples even if it's
lost.  Currenlty PERF_RECORD_LOST is generated for a ring-buffer which
might be shared with other events.  So it's hard to know per-event
lost count.

Add event->lost_samples field and PERF_FORMAT_LOST to retrieve it from
userspace.
Original-patch-by: Jiri Olsa <jolsa@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20220616180623.1358843-1-namhyung@kernel.org

include/linux/perf_event.h

@@ -759,6 +759,8 @@ struct perf_event {
 	struct pid_namespace		*ns;
 	u64				id;
 
+	atomic64_t			lost_samples;
+
 	u64				(*clock)(void);
 	perf_overflow_handler_t		overflow_handler;
 	void				*overflow_handler_context;

include/uapi/linux/perf_event.h

@@ -301,6 +301,7 @@ enum {
  *	  { u64		time_enabled; } && PERF_FORMAT_TOTAL_TIME_ENABLED
  *	  { u64		time_running; } && PERF_FORMAT_TOTAL_TIME_RUNNING
  *	  { u64		id;           } && PERF_FORMAT_ID
+ *	  { u64		lost;         } && PERF_FORMAT_LOST
  *	} && !PERF_FORMAT_GROUP
  *
  *	{ u64		nr;
@@ -308,6 +309,7 @@ enum {
  *	  { u64		time_running; } && PERF_FORMAT_TOTAL_TIME_RUNNING
  *	  { u64		value;
  *	    { u64	id;           } && PERF_FORMAT_ID
+ *	    { u64	lost;         } && PERF_FORMAT_LOST
  *	  }		cntr[nr];
  *	} && PERF_FORMAT_GROUP
  * };
@@ -317,8 +319,9 @@ enum perf_event_read_format {
 	PERF_FORMAT_TOTAL_TIME_RUNNING		= 1U << 1,
 	PERF_FORMAT_ID				= 1U << 2,
 	PERF_FORMAT_GROUP			= 1U << 3,
+	PERF_FORMAT_LOST			= 1U << 4,
 
-	PERF_FORMAT_MAX = 1U << 4,		/* non-ABI */
+	PERF_FORMAT_MAX = 1U << 5,		/* non-ABI */
 };
 
 #define PERF_ATTR_SIZE_VER0	64	/* sizeof first published struct */

kernel/events/core.c

@@ -1819,6 +1819,9 @@ static void __perf_event_read_size(struct perf_event *event, int nr_siblings)
 	if (event->attr.read_format & PERF_FORMAT_ID)
 		entry += sizeof(u64);
 
+	if (event->attr.read_format & PERF_FORMAT_LOST)
+		entry += sizeof(u64);
+
 	if (event->attr.read_format & PERF_FORMAT_GROUP) {
 		nr += nr_siblings;
 		size += sizeof(u64);
@@ -5260,11 +5263,15 @@ static int __perf_read_group_add(struct perf_event *leader,
 	values[n++] += perf_event_count(leader);
 	if (read_format & PERF_FORMAT_ID)
 		values[n++] = primary_event_id(leader);
+	if (read_format & PERF_FORMAT_LOST)
+		values[n++] = atomic64_read(&leader->lost_samples);
 
 	for_each_sibling_event(sub, leader) {
 		values[n++] += perf_event_count(sub);
 		if (read_format & PERF_FORMAT_ID)
 			values[n++] = primary_event_id(sub);
+		if (read_format & PERF_FORMAT_LOST)
+			values[n++] = atomic64_read(&sub->lost_samples);
 	}
 
 	raw_spin_unlock_irqrestore(&ctx->lock, flags);
@@ -5321,7 +5328,7 @@ static int perf_read_one(struct perf_event *event,
 				 u64 read_format, char __user *buf)
 {
 	u64 enabled, running;
-	u64 values[4];
+	u64 values[5];
 	int n = 0;
 
 	values[n++] = __perf_event_read_value(event, &enabled, &running);
@@ -5331,6 +5338,8 @@ static int perf_read_one(struct perf_event *event,
 		values[n++] = running;
 	if (read_format & PERF_FORMAT_ID)
 		values[n++] = primary_event_id(event);
+	if (read_format & PERF_FORMAT_LOST)
+		values[n++] = atomic64_read(&event->lost_samples);
 
 	if (copy_to_user(buf, values, n * sizeof(u64)))
 		return -EFAULT;
@@ -6858,7 +6867,7 @@ static void perf_output_read_one(struct perf_output_handle *handle,
 				 u64 enabled, u64 running)
 {
 	u64 read_format = event->attr.read_format;
-	u64 values[4];
+	u64 values[5];
 	int n = 0;
 
 	values[n++] = perf_event_count(event);
@@ -6872,6 +6881,8 @@ static void perf_output_read_one(struct perf_output_handle *handle,
 	}
 	if (read_format & PERF_FORMAT_ID)
 		values[n++] = primary_event_id(event);
+	if (read_format & PERF_FORMAT_LOST)
+		values[n++] = atomic64_read(&event->lost_samples);
 
 	__output_copy(handle, values, n * sizeof(u64));
 }
@@ -6882,7 +6893,7 @@ static void perf_output_read_group(struct perf_output_handle *handle,
 {
 	struct perf_event *leader = event->group_leader, *sub;
 	u64 read_format = event->attr.read_format;
-	u64 values[5];
+	u64 values[6];
 	int n = 0;
 
 	values[n++] = 1 + leader->nr_siblings;
@@ -6900,6 +6911,8 @@ static void perf_output_read_group(struct perf_output_handle *handle,
 	values[n++] = perf_event_count(leader);
 	if (read_format & PERF_FORMAT_ID)
 		values[n++] = primary_event_id(leader);
+	if (read_format & PERF_FORMAT_LOST)
+		values[n++] = atomic64_read(&leader->lost_samples);
 
 	__output_copy(handle, values, n * sizeof(u64));
 
@@ -6913,6 +6926,8 @@ static void perf_output_read_group(struct perf_output_handle *handle,
 		values[n++] = perf_event_count(sub);
 		if (read_format & PERF_FORMAT_ID)
 			values[n++] = primary_event_id(sub);
+		if (read_format & PERF_FORMAT_LOST)
+			values[n++] = atomic64_read(&sub->lost_samples);
 
 		__output_copy(handle, values, n * sizeof(u64));
 	}

kernel/events/ring_buffer.c

@@ -172,8 +172,10 @@ __perf_output_begin(struct perf_output_handle *handle,
 		goto out;
 
 	if (unlikely(rb->paused)) {
-		if (rb->nr_pages)
+		if (rb->nr_pages) {
 			local_inc(&rb->lost);
+			atomic64_inc(&event->lost_samples);
+		}
 		goto out;
 	}
 
@@ -254,6 +256,7 @@ __perf_output_begin(struct perf_output_handle *handle,
 
 fail:
 	local_inc(&rb->lost);
+	atomic64_inc(&event->lost_samples);
 	perf_output_put_handle(handle);
 out:
 	rcu_read_unlock();