Skip to contentSkip to conversation

perf: Do not POLLHUP event if it has children

PATCH In review Jiri Olsa 3 participants 6 messages Sep 12, 2014 – Sep 24, 2014 perf events

lore
Share & tools
lore

Jiri Olsa

author reviewer
#0 lore raw

PATCH 1/3 perf: Do not POLLHUP event if it has children

Currently we return POLLHUP in event polling if the monitored process is done, but we didn't consider possible children, that might be still running and producing data.

Before returning POLLHUP making sure that
1) the monitored task has exited and that
2) we don't have any children to monitor

Also adding parent wakeup when the child event is gone.

Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 kernel/events/core.c | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)
kernel/events/core.c +20 -1
index d8cb4d21a346..cf779eaaf624 100644
@@ -3586,6 +3586,19 @@ static int perf_event_read_one(struct perf_event *event,
 	return n * sizeof(u64);
 }
 
+static bool is_event_hup(struct perf_event *event)
+{
+	bool no_children;
+
+	if (event->state != PERF_EVENT_STATE_EXIT)
+		return false;
+
+	mutex_lock(&event->child_mutex);
+	no_children = list_empty(&event->child_list);
+	mutex_unlock(&event->child_mutex);
+	return no_children;
+}
+
 /*
  * Read the performance event - simple non blocking version for now
  */
@@ -3632,7 +3645,7 @@ static unsigned int perf_poll(struct file *file, poll_table *wait)
 
 	poll_wait(file, &event->waitq, wait);
 
-	if (event->state == PERF_EVENT_STATE_EXIT)
+	if (is_event_hup(event))
 		return events;
 
 	/*
@@ -7560,6 +7573,12 @@ static void sync_child_event(struct perf_event *child_event,
 	mutex_unlock(&parent_event->child_mutex);
 
 	/*
+	 * Make sure user/parent get notified, that we just
+	 * lost one event.
+	 */
+	perf_event_wakeup(parent_event);
+
+	/*
 	 * Release the parent event, if this was the last
 	 * reference to it.
 	 */
Signature · 1 line
1.8.3.1
author reviewer
#1 lore raw

PATCH 2/3 perf: Fix child event initial state setup

From: Jiri Olsa <[email protected]>

Currently we initialize the child event based on the original parent state. This is wrong, because the original parent event (and its state) is not related to current fork and also could be already gone.

We need to initialize the child state based on the immediate parent event state.

Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 kernel/events/core.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
kernel/events/core.c +2 -1
index cf779eaaf624..5a09a09b2983 100644
@@ -7798,6 +7798,7 @@ inherit_event(struct perf_event *parent_event,
 	      struct perf_event *group_leader,
 	      struct perf_event_context *child_ctx)
 {
+	enum perf_event_active_state parent_state = parent_event->state;
 	struct perf_event *child_event;
 	unsigned long flags;
 
@@ -7831,7 +7832,7 @@ inherit_event(struct perf_event *parent_event,
 	 * not its attr.disabled bit.  We hold the parent's mutex,
 	 * so we won't race with perf_event_{en, dis}able_family.
 	 */
-	if (parent_event->state >= PERF_EVENT_STATE_INACTIVE)
+	if (parent_state >= PERF_EVENT_STATE_INACTIVE)
 		child_event->state = PERF_EVENT_STATE_INACTIVE;
 	else
 		child_event->state = PERF_EVENT_STATE_OFF;
Signature · 1 line
1.8.3.1
#2 lore raw

[tip:perf/core] perf: Fix child event initial state setup

perf: Fix child event initial state setup
Commit-ID:  1929def9e609d1a8cdb1626d85eda3da66921a7d
Gitweb:     http://git.kernel.org/tip/1929def9e609d1a8cdb1626d85eda3da66921a7d
Author:     Jiri Olsa <[email protected]>
AuthorDate: Fri, 12 Sep 2014 13:18:27 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 24 Sep 2014 14:48:12 +0200

perf: Fix child event initial state setup

Currently we initialize the child event based on the original parent state. This is wrong, because the original parent event (and its state) is not related to current fork and also could be already gone.

We need to initialize the child state based on the immediate parent event state.

---
 kernel/events/core.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
kernel/events/core.c +2 -1
index 15e58d4..132524c 100644
@@ -7818,6 +7818,7 @@ inherit_event(struct perf_event *parent_event,
 	      struct perf_event *group_leader,
 	      struct perf_event_context *child_ctx)
 {
+	enum perf_event_active_state parent_state = parent_event->state;
 	struct perf_event *child_event;
 	unsigned long flags;
 
@@ -7851,7 +7852,7 @@ inherit_event(struct perf_event *parent_event,
 	 * not its attr.disabled bit.  We hold the parent's mutex,
 	 * so we won't race with perf_event_{en, dis}able_family.
 	 */
-	if (parent_event->state >= PERF_EVENT_STATE_INACTIVE)
+	if (parent_state >= PERF_EVENT_STATE_INACTIVE)
 		child_event->state = PERF_EVENT_STATE_INACTIVE;
 	else
 		child_event->state = PERF_EVENT_STATE_OFF;
author reviewer
#3 lore raw

PATCH 3/3 Revert "perf: Do not allow optimized switch for non-cloned events"

In reply to Jiri Olsa

From: Jiri Olsa <[email protected]>

This reverts commit 1f9a7268c67f0290837aada443d28fd953ddca90.

With the fix of the initial state for the cloned event we now correctly handle the error described in:
1f9a7268c67f perf: Do not allow optimized switch for non-cloned events
so we can revert it.

I made an automated test for this, but its not suitable for automated perf tests framework. It needs to be customized for each machine (the more cpu the higher numbers for GROUPS/WORKERS/BYTES) and it could take longer time to hit the issue. Adding at least a lkml link with test patch.

Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
---
 kernel/events/core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/events/core.c +1 -1
index 5a09a09b2983..532670fbfe12 100644
@@ -2374,7 +2374,7 @@ static void perf_event_context_sched_out(struct task_struct *task, int ctxn,
 	next_parent = rcu_dereference(next_ctx->parent_ctx);
 
 	/* If neither context have a parent context; they cannot be clones. */
-	if (!parent || !next_parent)
+	if (!parent && !next_parent)
 		goto unlock;
 
 	if (next_parent == ctx || next_ctx == parent || next_parent == parent) {
Signature · 1 line
1.8.3.1
maintainer
#4 lore raw
In reply to Jiri Olsa
Quoted 18 lines · Jiri Olsa

Currently we return POLLHUP in event polling if the monitored process is done, but we didn't consider possible children, that might be still running and producing data.

Before returning POLLHUP making sure that
1) the monitored task has exited and that
2) we don't have any children to monitor

Also adding parent wakeup when the child event is gone.

Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
Quoted 1 line

---

Thanks Jiri (for all three patches)!

#5 lore raw
In reply to Jiri Olsa
perf: Do not POLLHUP event if it has children
Commit-ID:  dc633982ff3f4fd74cdc11b5a6ae53d39a0b2451
Gitweb:     http://git.kernel.org/tip/dc633982ff3f4fd74cdc11b5a6ae53d39a0b2451
Author:     Jiri Olsa <[email protected]>
AuthorDate: Fri, 12 Sep 2014 13:18:26 +0200
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 24 Sep 2014 14:48:11 +0200

perf: Do not POLLHUP event if it has children

Currently we return POLLHUP in event polling if the monitored process is done, but we didn't consider possible children, that might be still running and producing data.

Before returning POLLHUP making sure that:

1) the monitored task has exited and that
2) we don't have any children to monitor

Also adding parent wakeup when the child event is gone.

Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Stephane Eranian <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Frederic Weisbecker <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc Stephane Eranian <[email protected]>
Cc Linus Torvalds <[email protected]>
---
 kernel/events/core.c | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)
kernel/events/core.c +20 -1
index 733c616..15e58d4 100644
@@ -3587,6 +3587,19 @@ static int perf_event_read_one(struct perf_event *event,
 	return n * sizeof(u64);
 }
 
+static bool is_event_hup(struct perf_event *event)
+{
+	bool no_children;
+
+	if (event->state != PERF_EVENT_STATE_EXIT)
+		return false;
+
+	mutex_lock(&event->child_mutex);
+	no_children = list_empty(&event->child_list);
+	mutex_unlock(&event->child_mutex);
+	return no_children;
+}
+
 /*
  * Read the performance event - simple non blocking version for now
  */
@@ -3632,7 +3645,7 @@ static unsigned int perf_poll(struct file *file, poll_table *wait)
 
 	poll_wait(file, &event->waitq, wait);
 
-	if (event->state == PERF_EVENT_STATE_EXIT)
+	if (is_event_hup(event))
 		return events;
 
 	/*
@@ -7580,6 +7593,12 @@ static void sync_child_event(struct perf_event *child_event,
 	mutex_unlock(&parent_event->child_mutex);
 
 	/*
+	 * Make sure user/parent get notified, that we just
+	 * lost one event.
+	 */
+	perf_event_wakeup(parent_event);
+
+	/*
 	 * Release the parent event, if this was the last
 	 * reference to it.
 	 */