Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1311502 > unrolled thread

[PATCH] perf core: Introduce new ioctl options to pause and resume ring buffer

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  [PATCH] perf core: Introduce new ioctl options to pause and resume ring buffer Wang Nan <wangnan0@huawei.com> - 2016-01-18 13:00 +0100
    Re: [PATCH] perf core: Introduce new ioctl options to pause and  resume ring buffer Peter Zijlstra <peterz@infradead.org> - 2016-01-18 13:10 +0100
      Re: [PATCH] perf core: Introduce new ioctl options to pause and resume  ring buffer "Wangnan (F)" <wangnan0@huawei.com> - 2016-01-19 04:00 +0100
      [PATCH 0/6] perf core: Read from overwrite ring buffer Wang Nan <wangnan0@huawei.com> - 2016-01-19 12:20 +0100
        [PATCH 1/6] perf core: Introduce new ioctl options to pause and resume ring buffer Wang Nan <wangnan0@huawei.com> - 2016-01-19 12:20 +0100
        [PATCH 6/6] perf/core: Put size of a sample at the end of it by PERF_SAMPLE_TAILSIZE Wang Nan <wangnan0@huawei.com> - 2016-01-19 12:20 +0100
        [PATCH 2/6] perf core: Set event's default overflow_handler Wang Nan <wangnan0@huawei.com> - 2016-01-19 12:20 +0100
        [PATCH 5/6] perf core: Reduce perf event output overhead by setting overwrite handler Wang Nan <wangnan0@huawei.com> - 2016-01-19 12:20 +0100
        [PATCH 4/6] perf core: Add backwork attribute to perf event Wang Nan <wangnan0@huawei.com> - 2016-01-19 12:20 +0100
        [PATCH 3/6] perf core: Prepare writing into ring buffer from end Wang Nan <wangnan0@huawei.com> - 2016-01-19 12:30 +0100
        Re: [PATCH 0/6] perf core: Read from overwrite ring buffer Namhyung Kim <namhyung@kernel.org> - 2016-01-19 15:10 +0100
          Re: [PATCH 0/6] perf core: Read from overwrite ring buffer pi3orama <pi3orama@163.com> - 2016-01-19 15:20 +0100
        Re: [PATCH 0/6] perf core: Read from overwrite ring buffer Alexei Starovoitov <alexei.starovoitov@gmail.com> - 2016-01-19 18:50 +0100
          Re: [PATCH 0/6] perf core: Read from overwrite ring buffer "Wangnan (F)" <wangnan0@huawei.com> - 2016-01-20 02:40 +0100
            Re: [PATCH 0/6] perf core: Read from overwrite ring buffer Alexei Starovoitov <alexei.starovoitov@gmail.com> - 2016-01-20 03:30 +0100
              Re: [PATCH 0/6] perf core: Read from overwrite ring buffer "Wangnan (F)" <wangnan0@huawei.com> - 2016-01-21 08:00 +0100
                Re: [PATCH 0/6] perf core: Read from overwrite ring buffer "Wangnan (F)" <wangnan0@huawei.com> - 2016-01-22 03:30 +0100
                  Re: [PATCH 0/6] perf core: Read from overwrite ring buffer Alexei Starovoitov <alexei.starovoitov@gmail.com> - 2016-01-22 04:30 +0100
                    Re: [PATCH 0/6] perf core: Read from overwrite ring buffer "Wangnan (F)" <wangnan0@huawei.com> - 2016-01-22 05:50 +0100

#1311502 — [PATCH] perf core: Introduce new ioctl options to pause and resume ring buffer

Add an extra ioctl() to pause/resume ring-buffer output.

In some situations we want to read from ring buffer only when we
ensure nothing can write to the ring buffer during reading. Without
this patch we have to turn off all events attached to this ring buffer.
This patch is for supporting overwritable ring buffer with TAILSIZE
selected.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 include/uapi/linux/perf_event.h |  2 ++
 kernel/events/core.c            | 14 ++++++++++++++
 kernel/events/internal.h        | 11 +++++++++++
 kernel/events/ring_buffer.c     |  4 +++-
 4 files changed, 30 insertions(+), 1 deletion(-)

diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index 4e8dde8..9508070 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -402,6 +402,8 @@ struct perf_event_attr {
 #define PERF_EVENT_IOC_SET_FILTER	_IOW('$', 6, char *)
 #define PERF_EVENT_IOC_ID		_IOR('$', 7, __u64 *)
 #define PERF_EVENT_IOC_SET_BPF		_IOW('$', 8, __u32)
+#define PERF_EVENT_IOC_PAUSE_OUTPUT	_IO ('$', 9)
+#define PERF_EVENT_IOC_RESUME_OUTPUT	_IO ('$', 10)
 
 enum perf_event_ioc_flags {
 	PERF_IOC_FLAG_GROUP		= 1U << 0,
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 2d59b59..d5a0c34 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -4241,6 +4241,20 @@ static long _perf_ioctl(struct perf_event *event, unsigned int cmd, unsigned lon
 	case PERF_EVENT_IOC_SET_BPF:
 		return perf_event_set_bpf_prog(event, arg);
 
+	case PERF_EVENT_IOC_PAUSE_OUTPUT:
+	case PERF_EVENT_IOC_RESUME_OUTPUT: {
+		struct ring_buffer *rb;
+
+		rcu_read_lock();
+		rb = rcu_dereference(event->rb);
+		if (!event->rb) {
+			rcu_read_unlock();
+			return -EINVAL;
+		}
+		rb_toggle_paused(rb, cmd == PERF_EVENT_IOC_PAUSE_OUTPUT);
+		rcu_read_unlock();
+		return 0;
+	}
 	default:
 		return -ENOTTY;
 	}
diff --git a/kernel/events/internal.h b/kernel/events/internal.h
index 2bbad9c..6a93d1b 100644
--- a/kernel/events/internal.h
+++ b/kernel/events/internal.h
@@ -18,6 +18,7 @@ struct ring_buffer {
 #endif
 	int				nr_pages;	/* nr of data pages  */
 	int				overwrite;	/* can overwrite itself */
+	int				paused;		/* can write into ring buffer */
 
 	atomic_t			poll;		/* POLL_ for wakeups */
 
@@ -65,6 +66,16 @@ static inline void rb_free_rcu(struct rcu_head *rcu_head)
 	rb_free(rb);
 }
 
+static inline void
+rb_toggle_paused(struct ring_buffer *rb,
+		 bool pause)
+{
+	if (!pause && rb->nr_pages)
+		rb->paused = 0;
+	else
+		rb->paused = 1;
+}
+
 extern struct ring_buffer *
 rb_alloc(int nr_pages, long watermark, int cpu, int flags);
 extern void perf_event_wakeup(struct perf_event *event);
diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index 5f8bd89..11a1676 100644
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -125,7 +125,7 @@ int perf_output_begin(struct perf_output_handle *handle,
 	if (unlikely(!rb))
 		goto out;
 
-	if (unlikely(!rb->nr_pages))
+	if (unlikely(rb->paused))
 		goto out;
 
 	handle->rb    = rb;
@@ -245,6 +245,8 @@ ring_buffer_init(struct ring_buffer *rb, long watermark, int flags)
 	INIT_LIST_HEAD(&rb->event_list);
 	spin_lock_init(&rb->event_lock);
 	init_irq_work(&rb->irq_work, rb_irq_work);
+
+	rb->paused = rb->nr_pages ? 0 : 1;
 }
 
 static void ring_buffer_put_async(struct ring_buffer *rb)
-- 
1.8.3.4

[toc] | [next] | [standalone]

#1311507 — Re: [PATCH] perf core: Introduce new ioctl options to pause and resume ring buffer

On Mon, Jan 18, 2016 at 11:52:01AM +0000, Wang Nan wrote:

> +#define PERF_EVENT_IOC_PAUSE_OUTPUT	_IO ('$', 9)
> +#define PERF_EVENT_IOC_RESUME_OUTPUT	_IO ('$', 10)

Would not a single IOCTL with a 'boolean' parameter make more sense?

> +++ b/kernel/events/ring_buffer.c
> @@ -125,7 +125,7 @@ int perf_output_begin(struct perf_output_handle *handle,
>  	if (unlikely(!rb))
>  		goto out;
>  
> -	if (unlikely(!rb->nr_pages))
> +	if (unlikely(rb->paused))
>  		goto out;

Should we increment rb->lost in this case?

[toc] | [prev] | [next] | [standalone]

#1311913 — Re: [PATCH] perf core: Introduce new ioctl options to pause and resume ring buffer

On 2016/1/18 20:02, Peter Zijlstra wrote:
> On Mon, Jan 18, 2016 at 11:52:01AM +0000, Wang Nan wrote:
>
>> +#define PERF_EVENT_IOC_PAUSE_OUTPUT	_IO ('$', 9)
>> +#define PERF_EVENT_IOC_RESUME_OUTPUT	_IO ('$', 10)
> Would not a single IOCTL with a 'boolean' parameter make more sense?

Good suggestion.

>> +++ b/kernel/events/ring_buffer.c
>> @@ -125,7 +125,7 @@ int perf_output_begin(struct perf_output_handle *handle,
>>   	if (unlikely(!rb))
>>   		goto out;
>>   
>> -	if (unlikely(!rb->nr_pages))
>> +	if (unlikely(rb->paused))
>>   		goto out;
> Should we increment rb->lost in this case?

Not sure about this. The ring buffer is paused deliberately, shall we 
consider the
events we miss as losted events? However I'll try it in next version.

Thank you.

[toc] | [prev] | [next] | [standalone]

#1312125 — [PATCH 0/6] perf core: Read from overwrite ring buffer

This patchset introduces two methods to support reading from overwrite.

 1) Tailsize: write the size of an event at the end of it
 2) Backward writing: write the ring buffer from the end of it to the
    beginning.

Patch 1/6 introduces a new ioctl operation to pause and resume ring
buffer since reading from a overwrite ring buffer is not reliable.

To reduce overhead as much as possible, force setting overflow_handler
and create specific function for backward writing and onward writing.

Cc: He Kuang <hekuang@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com

Wang Nan (6):
  perf core: Introduce new ioctl options to pause and resume ring buffer
  perf core: Set event's default overflow_handler
  perf core: Prepare writing into ring buffer from end
  perf core: Add backwork attribute to perf event
  perf core: Reduce perf event output overhead by setting overwrite
    handler
  perf/core: Put size of a sample at the end of it by
    PERF_SAMPLE_TAILSIZE

 include/linux/perf_event.h      |  39 +++++++---
 include/uapi/linux/perf_event.h |   7 +-
 kernel/events/core.c            | 155 +++++++++++++++++++++++++++++++---------
 kernel/events/internal.h        |  11 +++
 kernel/events/ring_buffer.c     |  65 ++++++++++++++---
 5 files changed, 223 insertions(+), 54 deletions(-)

-- 
1.8.3.4

[toc] | [prev] | [next] | [standalone]

#1312128 — [PATCH 1/6] perf core: Introduce new ioctl options to pause and resume ring buffer

Add new ioctl() to pause/resume ring-buffer output.

In some situations we want to read from ring buffer only when we
ensure nothing can write to the ring buffer during reading. Without
this patch we have to turn off all events attached to this ring buffer.
This patch is for supporting overwritable ring buffer with TAILSIZE
selected.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 include/uapi/linux/perf_event.h |  1 +
 kernel/events/core.c            | 13 +++++++++++++
 kernel/events/internal.h        | 11 +++++++++++
 kernel/events/ring_buffer.c     |  7 ++++++-
 4 files changed, 31 insertions(+), 1 deletion(-)

diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index 1afe962..2c7f00c 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -401,6 +401,7 @@ struct perf_event_attr {
 #define PERF_EVENT_IOC_SET_FILTER	_IOW('$', 6, char *)
 #define PERF_EVENT_IOC_ID		_IOR('$', 7, __u64 *)
 #define PERF_EVENT_IOC_SET_BPF		_IOW('$', 8, __u32)
+#define PERF_EVENT_IOC_PAUSE_OUTPUT	_IO ('$', 9)
 
 enum perf_event_ioc_flags {
 	PERF_IOC_FLAG_GROUP		= 1U << 0,
diff --git a/kernel/events/core.c b/kernel/events/core.c
index bf82441..9e9c84da 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -4241,6 +4241,19 @@ static long _perf_ioctl(struct perf_event *event, unsigned int cmd, unsigned lon
 	case PERF_EVENT_IOC_SET_BPF:
 		return perf_event_set_bpf_prog(event, arg);
 
+	case PERF_EVENT_IOC_PAUSE_OUTPUT: {
+		struct ring_buffer *rb;
+
+		rcu_read_lock();
+		rb = rcu_dereference(event->rb);
+		if (!event->rb) {
+			rcu_read_unlock();
+			return -EINVAL;
+		}
+		rb_toggle_paused(rb, !!arg);
+		rcu_read_unlock();
+		return 0;
+	}
 	default:
 		return -ENOTTY;
 	}
diff --git a/kernel/events/internal.h b/kernel/events/internal.h
index 2bbad9c..6a93d1b 100644
--- a/kernel/events/internal.h
+++ b/kernel/events/internal.h
@@ -18,6 +18,7 @@ struct ring_buffer {
 #endif
 	int				nr_pages;	/* nr of data pages  */
 	int				overwrite;	/* can overwrite itself */
+	int				paused;		/* can write into ring buffer */
 
 	atomic_t			poll;		/* POLL_ for wakeups */
 
@@ -65,6 +66,16 @@ static inline void rb_free_rcu(struct rcu_head *rcu_head)
 	rb_free(rb);
 }
 
+static inline void
+rb_toggle_paused(struct ring_buffer *rb,
+		 bool pause)
+{
+	if (!pause && rb->nr_pages)
+		rb->paused = 0;
+	else
+		rb->paused = 1;
+}
+
 extern struct ring_buffer *
 rb_alloc(int nr_pages, long watermark, int cpu, int flags);
 extern void perf_event_wakeup(struct perf_event *event);
diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index adfdc05..9f1a93f 100644
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -125,8 +125,11 @@ int perf_output_begin(struct perf_output_handle *handle,
 	if (unlikely(!rb))
 		goto out;
 
-	if (unlikely(!rb->nr_pages))
+	if (unlikely(rb->paused)) {
+		if (rb->nr_pages)
+			local_inc(&rb->lost);
 		goto out;
+	}
 
 	handle->rb    = rb;
 	handle->event = event;
@@ -244,6 +247,8 @@ ring_buffer_init(struct ring_buffer *rb, long watermark, int flags)
 	INIT_LIST_HEAD(&rb->event_list);
 	spin_lock_init(&rb->event_lock);
 	init_irq_work(&rb->irq_work, rb_irq_work);
+
+	rb->paused = rb->nr_pages ? 0 : 1;
 }
 
 static void ring_buffer_put_async(struct ring_buffer *rb)
-- 
1.8.3.4

[toc] | [prev] | [next] | [standalone]

#1312129 — [PATCH 6/6] perf/core: Put size of a sample at the end of it by PERF_SAMPLE_TAILSIZE

This patch introduces a PERF_SAMPLE_TAILSIZE flag which allows a size
field attached at the end of a sample. The idea comes from [1] that,
with tie size at tail of an event, it is possible for user program who
read from the ring buffer parse events backward.

For example:

   head
    |
    V
 +--+---+-------+----------+------+---+
 |E6|...|   B  8|   C    11|  D  7|E..|
 +--+---+-------+----------+------+---+

In this case, from the 'head' pointer provided by kernel, user program
can first see '6' by (*(head - sizeof(u64))), then it can get the start
pointer of record 'E', then it can read size and find start position
of record D, C, B in similar way.

The implementation is easy: adding a PERF_SAMPLE_TAILSIZE flag, makes
perf_output_sample() output size at the end of a sample.

Following things are done for ensure the ring buffer is safe for
backward parsing:

 - Don't allow two events with different PERF_SAMPLE_TAILSIZE setting
   set their output to each other;

 - For non-sample events, also output tailsize if required.

This patch has a limitation for perf:

Before reading such ring buffer, perf must ensure all events which may
output to it is already stopped, so the 'head' pointer it get is the
end of the last record.

[1] http://lkml.kernel.org/g/1449063499-236703-1-git-send-email-wangnan0@huawei.com

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Yunlong Song <yunlong.song@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 include/linux/perf_event.h      | 17 ++++++---
 include/uapi/linux/perf_event.h |  3 +-
 kernel/events/core.c            | 82 +++++++++++++++++++++++++++++------------
 kernel/events/ring_buffer.c     |  7 ++--
 4 files changed, 75 insertions(+), 34 deletions(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index c0335b9..7c70d4b 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -841,13 +841,13 @@ extern void perf_event_output(struct perf_event *event,
 			      struct pt_regs *regs);
 
 extern void
-perf_event_header__init_id(struct perf_event_header *header,
-			   struct perf_sample_data *data,
-			   struct perf_event *event);
+perf_event_header__init_extra(struct perf_event_header *header,
+			      struct perf_sample_data *data,
+			      struct perf_event *event);
 extern void
-perf_event__output_id_sample(struct perf_event *event,
-			     struct perf_output_handle *handle,
-			     struct perf_sample_data *sample);
+perf_event__output_extra(struct perf_event *event, u64 evt_size,
+			 struct perf_output_handle *handle,
+			 struct perf_sample_data *sample);
 
 extern void
 perf_log_lost_samples(struct perf_event *event, u64 lost);
@@ -1043,6 +1043,11 @@ static inline bool is_write_backward(struct perf_event *event)
 	return !!event->attr.write_backward;
 }
 
+static inline bool has_tailsize(struct perf_event *event)
+{
+	return !!(event->attr.sample_type & PERF_SAMPLE_TAILSIZE);
+}
+
 extern int perf_output_begin(struct perf_output_handle *handle,
 			     struct perf_event *event, unsigned int size);
 extern int perf_output_begin_onward(struct perf_output_handle *handle,
diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index 598b9b0..f0cad26 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -139,8 +139,9 @@ enum perf_event_sample_format {
 	PERF_SAMPLE_IDENTIFIER			= 1U << 16,
 	PERF_SAMPLE_TRANSACTION			= 1U << 17,
 	PERF_SAMPLE_REGS_INTR			= 1U << 18,
+	PERF_SAMPLE_TAILSIZE			= 1U << 19,
 
-	PERF_SAMPLE_MAX = 1U << 19,		/* non-ABI */
+	PERF_SAMPLE_MAX = 1U << 20,		/* non-ABI */
 };
 
 /*
diff --git a/kernel/events/core.c b/kernel/events/core.c
index fa32d8c..d8bb92e 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -5141,12 +5141,14 @@ static void __perf_event_header__init_id(struct perf_event_header *header,
 	}
 }
 
-void perf_event_header__init_id(struct perf_event_header *header,
-				struct perf_sample_data *data,
-				struct perf_event *event)
+void perf_event_header__init_extra(struct perf_event_header *header,
+				   struct perf_sample_data *data,
+				   struct perf_event *event)
 {
 	if (event->attr.sample_id_all)
 		__perf_event_header__init_id(header, data, event);
+	if (has_tailsize(event))
+		header->size += sizeof(u64);
 }
 
 static void __perf_event__output_id_sample(struct perf_output_handle *handle,
@@ -5173,12 +5175,14 @@ static void __perf_event__output_id_sample(struct perf_output_handle *handle,
 		perf_output_put(handle, data->id);
 }
 
-void perf_event__output_id_sample(struct perf_event *event,
-				  struct perf_output_handle *handle,
-				  struct perf_sample_data *sample)
+void perf_event__output_extra(struct perf_event *event, u64 evt_size,
+			      struct perf_output_handle *handle,
+			      struct perf_sample_data *sample)
 {
 	if (event->attr.sample_id_all)
 		__perf_event__output_id_sample(handle, sample);
+	if (has_tailsize(event))
+		perf_output_put(handle, evt_size);
 }
 
 static void perf_output_read_one(struct perf_output_handle *handle,
@@ -5420,6 +5424,13 @@ void perf_output_sample(struct perf_output_handle *handle,
 		}
 	}
 
+	/* Should be the last one */
+	if (sample_type & PERF_SAMPLE_TAILSIZE) {
+		u64 evt_size = header->size;
+
+		perf_output_put(handle, evt_size);
+	}
+
 	if (!event->attr.watermark) {
 		int wakeup_events = event->attr.wakeup_events;
 
@@ -5539,6 +5550,9 @@ void perf_prepare_sample(struct perf_event_header *header,
 
 		header->size += size;
 	}
+
+	if (sample_type & PERF_SAMPLE_TAILSIZE)
+		header->size += sizeof(u64);
 }
 
 static void __always_inline
@@ -5620,14 +5634,15 @@ perf_event_read_event(struct perf_event *event,
 	};
 	int ret;
 
-	perf_event_header__init_id(&read_event.header, &sample, event);
+	perf_event_header__init_extra(&read_event.header, &sample, event);
 	ret = perf_output_begin(&handle, event, read_event.header.size);
 	if (ret)
 		return;
 
 	perf_output_put(&handle, read_event);
 	perf_output_read(&handle, event);
-	perf_event__output_id_sample(event, &handle, &sample);
+	perf_event__output_extra(event, read_event.header.size,
+				 &handle, &sample);
 
 	perf_output_end(&handle);
 }
@@ -5739,7 +5754,7 @@ static void perf_event_task_output(struct perf_event *event,
 	if (!perf_event_task_match(event))
 		return;
 
-	perf_event_header__init_id(&task_event->event_id.header, &sample, event);
+	perf_event_header__init_extra(&task_event->event_id.header, &sample, event);
 
 	ret = perf_output_begin(&handle, event,
 				task_event->event_id.header.size);
@@ -5756,7 +5771,9 @@ static void perf_event_task_output(struct perf_event *event,
 
 	perf_output_put(&handle, task_event->event_id);
 
-	perf_event__output_id_sample(event, &handle, &sample);
+	perf_event__output_extra(event,
+				 task_event->event_id.header.size,
+				 &handle, &sample);
 
 	perf_output_end(&handle);
 out:
@@ -5835,7 +5852,7 @@ static void perf_event_comm_output(struct perf_event *event,
 	if (!perf_event_comm_match(event))
 		return;
 
-	perf_event_header__init_id(&comm_event->event_id.header, &sample, event);
+	perf_event_header__init_extra(&comm_event->event_id.header, &sample, event);
 	ret = perf_output_begin(&handle, event,
 				comm_event->event_id.header.size);
 
@@ -5849,7 +5866,8 @@ static void perf_event_comm_output(struct perf_event *event,
 	__output_copy(&handle, comm_event->comm,
 				   comm_event->comm_size);
 
-	perf_event__output_id_sample(event, &handle, &sample);
+	perf_event__output_extra(event, comm_event->event_id.header.size,
+				 &handle, &sample);
 
 	perf_output_end(&handle);
 out:
@@ -5958,7 +5976,7 @@ static void perf_event_mmap_output(struct perf_event *event,
 		mmap_event->event_id.header.size += sizeof(mmap_event->flags);
 	}
 
-	perf_event_header__init_id(&mmap_event->event_id.header, &sample, event);
+	perf_event_header__init_extra(&mmap_event->event_id.header, &sample, event);
 	ret = perf_output_begin(&handle, event,
 				mmap_event->event_id.header.size);
 	if (ret)
@@ -5981,7 +5999,8 @@ static void perf_event_mmap_output(struct perf_event *event,
 	__output_copy(&handle, mmap_event->file_name,
 				   mmap_event->file_size);
 
-	perf_event__output_id_sample(event, &handle, &sample);
+	perf_event__output_extra(event, mmap_event->event_id.header.size,
+				 &handle, &sample);
 
 	perf_output_end(&handle);
 out:
@@ -6164,14 +6183,15 @@ void perf_event_aux_event(struct perf_event *event, unsigned long head,
 	};
 	int ret;
 
-	perf_event_header__init_id(&rec.header, &sample, event);
+	perf_event_header__init_extra(&rec.header, &sample, event);
 	ret = perf_output_begin(&handle, event, rec.header.size);
 
 	if (ret)
 		return;
 
 	perf_output_put(&handle, rec);
-	perf_event__output_id_sample(event, &handle, &sample);
+	perf_event__output_extra(event, rec.header.size,
+				 &handle, &sample);
 
 	perf_output_end(&handle);
 }
@@ -6197,7 +6217,7 @@ void perf_log_lost_samples(struct perf_event *event, u64 lost)
 		.lost		= lost,
 	};
 
-	perf_event_header__init_id(&lost_samples_event.header, &sample, event);
+	perf_event_header__init_extra(&lost_samples_event.header, &sample, event);
 
 	ret = perf_output_begin(&handle, event,
 				lost_samples_event.header.size);
@@ -6205,7 +6225,8 @@ void perf_log_lost_samples(struct perf_event *event, u64 lost)
 		return;
 
 	perf_output_put(&handle, lost_samples_event);
-	perf_event__output_id_sample(event, &handle, &sample);
+	perf_event__output_extra(event, lost_samples_event.header.size,
+				 &handle, &sample);
 	perf_output_end(&handle);
 }
 
@@ -6252,7 +6273,7 @@ static void perf_event_switch_output(struct perf_event *event, void *data)
 					perf_event_tid(event, se->next_prev);
 	}
 
-	perf_event_header__init_id(&se->event_id.header, &sample, event);
+	perf_event_header__init_extra(&se->event_id.header, &sample, event);
 
 	ret = perf_output_begin(&handle, event, se->event_id.header.size);
 	if (ret)
@@ -6263,7 +6284,8 @@ static void perf_event_switch_output(struct perf_event *event, void *data)
 	else
 		perf_output_put(&handle, se->event_id);
 
-	perf_event__output_id_sample(event, &handle, &sample);
+	perf_event__output_extra(event, se->event_id.header.size,
+				 &handle, &sample);
 
 	perf_output_end(&handle);
 }
@@ -6323,7 +6345,7 @@ static void perf_log_throttle(struct perf_event *event, int enable)
 	if (enable)
 		throttle_event.header.type = PERF_RECORD_UNTHROTTLE;
 
-	perf_event_header__init_id(&throttle_event.header, &sample, event);
+	perf_event_header__init_extra(&throttle_event.header, &sample, event);
 
 	ret = perf_output_begin(&handle, event,
 				throttle_event.header.size);
@@ -6331,7 +6353,8 @@ static void perf_log_throttle(struct perf_event *event, int enable)
 		return;
 
 	perf_output_put(&handle, throttle_event);
-	perf_event__output_id_sample(event, &handle, &sample);
+	perf_event__output_extra(event, throttle_event.header.size,
+				 &handle, &sample);
 	perf_output_end(&handle);
 }
 
@@ -6359,14 +6382,15 @@ static void perf_log_itrace_start(struct perf_event *event)
 	rec.pid	= perf_event_pid(event, current);
 	rec.tid	= perf_event_tid(event, current);
 
-	perf_event_header__init_id(&rec.header, &sample, event);
+	perf_event_header__init_extra(&rec.header, &sample, event);
 	ret = perf_output_begin(&handle, event, rec.header.size);
 
 	if (ret)
 		return;
 
 	perf_output_put(&handle, rec);
-	perf_event__output_id_sample(event, &handle, &sample);
+	perf_event__output_extra(event, rec.header.size,
+				 &handle, &sample);
 
 	perf_output_end(&handle);
 }
@@ -8151,6 +8175,16 @@ perf_event_set_output(struct perf_event *event, struct perf_event *output_event)
 	    event->pmu != output_event->pmu)
 		goto out;
 
+	/*
+	 * Don't allow mixed tailsize setting since the resuling
+	 * ringbuffer would unable to be parsed backward.
+	 *
+	 * '!=' is safe because has_tailsize() returns bool, two differnt
+	 * non-zero values would be treated as equal (both true).
+	 */
+	if (has_tailsize(event) != has_tailsize(output_event))
+		goto out;
+
 set:
 	mutex_lock(&event->mmap_mutex);
 	/* Can't redirect output if we've got an active mmap() */
diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index 4b0ef33..5cb098e 100644
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -208,10 +208,11 @@ __perf_output_begin(struct perf_output_handle *handle,
 		lost_event.id          = event->id;
 		lost_event.lost        = local_xchg(&rb->lost, 0);
 
-		perf_event_header__init_id(&lost_event.header,
-					   &sample_data, event);
+		perf_event_header__init_extra(&lost_event.header,
+					      &sample_data, event);
 		perf_output_put(handle, lost_event);
-		perf_event__output_id_sample(event, handle, &sample_data);
+		perf_event__output_extra(event, lost_event.header.type,
+					 handle, &sample_data);
 	}
 
 	return 0;
-- 
1.8.3.4

[toc] | [prev] | [next] | [standalone]

#1312131 — [PATCH 2/6] perf core: Set event's default overflow_handler

Set a default event->overflow_handler in perf_event_alloc() so don't
need checking event->overflow_handler in __perf_event_overflow().
Following commits can give a different default overflow_handler.

No extra performance introduced into hot path because in the original
code we still need reading this handler from memory. A conditional branch
is avoided so actually we remove some instructions.

Initial idea comes from Peter at [1].

[1] http://lkml.kernel.org/r/20130708121557.GA17211@twins.programming.kicks-ass.net

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 kernel/events/core.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 9e9c84da..f79c4be 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -6402,10 +6402,7 @@ static int __perf_event_overflow(struct perf_event *event,
 		irq_work_queue(&event->pending);
 	}
 
-	if (event->overflow_handler)
-		event->overflow_handler(event, data, regs);
-	else
-		perf_event_output(event, data, regs);
+	event->overflow_handler(event, data, regs);
 
 	if (*perf_event_fasync(event) && event->pending_kill) {
 		event->pending_wakeup = 1;
@@ -7874,8 +7871,13 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
 		context = parent_event->overflow_handler_context;
 	}
 
-	event->overflow_handler	= overflow_handler;
-	event->overflow_handler_context = context;
+	if (overflow_handler) {
+		event->overflow_handler	= overflow_handler;
+		event->overflow_handler_context = context;
+	} else {
+		event->overflow_handler = perf_event_output;
+		event->overflow_handler_context = NULL;
+	}
 
 	perf_event__state_init(event);
 
-- 
1.8.3.4

[toc] | [prev] | [next] | [standalone]

#1312132 — [PATCH 5/6] perf core: Reduce perf event output overhead by setting overwrite handler

By creating onward and backward specific overflow handler and setting
them according to event's backward setting, normal sampling events
don't need to check backward setting of an event any more.

This is the last patch of backward writing patchset. After this patch,
there's no extra overhead introduced to the fast path of sampling
output.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 include/linux/perf_event.h  | 17 +++++++++++++++--
 kernel/events/core.c        | 41 ++++++++++++++++++++++++++++++++++++-----
 kernel/events/ring_buffer.c | 12 ++++++++++++
 3 files changed, 63 insertions(+), 7 deletions(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 54c3fb2..c0335b9 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -830,9 +830,15 @@ extern int perf_event_overflow(struct perf_event *event,
 				 struct perf_sample_data *data,
 				 struct pt_regs *regs);
 
+extern void perf_event_output_onward(struct perf_event *event,
+				     struct perf_sample_data *data,
+				     struct pt_regs *regs);
+extern void perf_event_output_backward(struct perf_event *event,
+				       struct perf_sample_data *data,
+				       struct pt_regs *regs);
 extern void perf_event_output(struct perf_event *event,
-				struct perf_sample_data *data,
-				struct pt_regs *regs);
+			      struct perf_sample_data *data,
+			      struct pt_regs *regs);
 
 extern void
 perf_event_header__init_id(struct perf_event_header *header,
@@ -1039,6 +1045,13 @@ static inline bool is_write_backward(struct perf_event *event)
 
 extern int perf_output_begin(struct perf_output_handle *handle,
 			     struct perf_event *event, unsigned int size);
+extern int perf_output_begin_onward(struct perf_output_handle *handle,
+				    struct perf_event *event,
+				    unsigned int size);
+extern int perf_output_begin_backward(struct perf_output_handle *handle,
+				      struct perf_event *event,
+				      unsigned int size);
+
 extern void perf_output_end(struct perf_output_handle *handle);
 extern unsigned int perf_output_copy(struct perf_output_handle *handle,
 			     const void *buf, unsigned int len);
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 8ad22a5..fa32d8c 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -5541,9 +5541,13 @@ void perf_prepare_sample(struct perf_event_header *header,
 	}
 }
 
-void perf_event_output(struct perf_event *event,
-			struct perf_sample_data *data,
-			struct pt_regs *regs)
+static void __always_inline
+__perf_event_output(struct perf_event *event,
+		    struct perf_sample_data *data,
+		    struct pt_regs *regs,
+		    int (*output_begin)(struct perf_output_handle *,
+			    		struct perf_event *,
+					unsigned int))
 {
 	struct perf_output_handle handle;
 	struct perf_event_header header;
@@ -5553,7 +5557,7 @@ void perf_event_output(struct perf_event *event,
 
 	perf_prepare_sample(&header, data, event, regs);
 
-	if (perf_output_begin(&handle, event, header.size))
+	if (output_begin(&handle, event, header.size))
 		goto exit;
 
 	perf_output_sample(&handle, &header, data, event);
@@ -5564,6 +5568,30 @@ exit:
 	rcu_read_unlock();
 }
 
+void
+perf_event_output_onward(struct perf_event *event,
+			 struct perf_sample_data *data,
+			 struct pt_regs *regs)
+{
+	__perf_event_output(event, data, regs, perf_output_begin_onward);
+}
+
+void
+perf_event_output_backward(struct perf_event *event,
+			   struct perf_sample_data *data,
+			   struct pt_regs *regs)
+{
+	__perf_event_output(event, data, regs, perf_output_begin_backward);
+}
+
+void
+perf_event_output(struct perf_event *event,
+		  struct perf_sample_data *data,
+		  struct pt_regs *regs)
+{
+	__perf_event_output(event, data, regs, perf_output_begin);
+}
+
 /*
  * read event_id
  */
@@ -7874,8 +7902,11 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
 	if (overflow_handler) {
 		event->overflow_handler	= overflow_handler;
 		event->overflow_handler_context = context;
+	} else if (is_write_backward(event)){
+		event->overflow_handler = perf_event_output_backward;
+		event->overflow_handler_context = NULL;
 	} else {
-		event->overflow_handler = perf_event_output;
+		event->overflow_handler = perf_event_output_onward;
 		event->overflow_handler_context = NULL;
 	}
 
diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index 1372427..4b0ef33 100644
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -225,6 +225,18 @@ out:
 	return -ENOSPC;
 }
 
+int perf_output_begin_onward(struct perf_output_handle *handle,
+			     struct perf_event *event, unsigned int size)
+{
+	return __perf_output_begin(handle, event, size, false);
+}
+
+int perf_output_begin_backward(struct perf_output_handle *handle,
+			       struct perf_event *event, unsigned int size)
+{
+	return __perf_output_begin(handle, event, size, true);
+}
+
 int perf_output_begin(struct perf_output_handle *handle,
 		      struct perf_event *event, unsigned int size)
 {
-- 
1.8.3.4

[toc] | [prev] | [next] | [standalone]

#1312133 — [PATCH 4/6] perf core: Add backwork attribute to perf event

In perf_event_attr a new bit 'write_backward' is appended to indicate
this event should write ring buffer from its end to beginning.

In perf_output_begin(), prepare ring buffer according this bit.

This patch introduces small overhead into perf_output_begin():
an extra memory read and a conditional branch. Further patch can remove
this overhead using custom output handler.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 include/linux/perf_event.h      | 5 +++++
 include/uapi/linux/perf_event.h | 3 ++-
 kernel/events/core.c            | 7 +++++++
 kernel/events/ring_buffer.c     | 2 ++
 4 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index f9828a4..54c3fb2 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -1032,6 +1032,11 @@ static inline bool has_aux(struct perf_event *event)
 	return event->pmu->setup_aux;
 }
 
+static inline bool is_write_backward(struct perf_event *event)
+{
+	return !!event->attr.write_backward;
+}
+
 extern int perf_output_begin(struct perf_output_handle *handle,
 			     struct perf_event *event, unsigned int size);
 extern void perf_output_end(struct perf_output_handle *handle);
diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index 2c7f00c..598b9b0 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -340,7 +340,8 @@ struct perf_event_attr {
 				comm_exec      :  1, /* flag comm events that are due to an exec */
 				use_clockid    :  1, /* use @clockid for time fields */
 				context_switch :  1, /* context switch data */
-				__reserved_1   : 37;
+				write_backward :  1, /* Write ring buffer from end to beginning */
+				__reserved_1   : 36;
 
 	union {
 		__u32		wakeup_events;	  /* wakeup every n events */
diff --git a/kernel/events/core.c b/kernel/events/core.c
index f79c4be..8ad22a5 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -8107,6 +8107,13 @@ perf_event_set_output(struct perf_event *event, struct perf_event *output_event)
 		goto out;
 
 	/*
+	 * Either writing ring buffer from beginning or from end.
+	 * Mixing is not allowed.
+	 */
+	if (is_write_backward(output_event) != is_write_backward(event))
+		goto out;
+
+	/*
 	 * If both events generate aux data, they must be on the same PMU
 	 */
 	if (has_aux(event) && has_aux(output_event) &&
diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index bbc3bc6..1372427 100644
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -228,6 +228,8 @@ out:
 int perf_output_begin(struct perf_output_handle *handle,
 		      struct perf_event *event, unsigned int size)
 {
+	if (unlikely(is_write_backward(event)))
+		return __perf_output_begin(handle, event, size, true);
 	return __perf_output_begin(handle, event, size, false);
 }
 
-- 
1.8.3.4

[toc] | [prev] | [next] | [standalone]

#1312136 — [PATCH 3/6] perf core: Prepare writing into ring buffer from end

Convert perf_output_begin to __perf_output_begin and make the later
function able to write records from the end of the ring buffer.
Following commits will utilize the 'backward' flag.

This patch doesn't introduce any extra performance overhead since we
use always_inline.

Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
---
 kernel/events/ring_buffer.c | 37 +++++++++++++++++++++++++++++++------
 1 file changed, 31 insertions(+), 6 deletions(-)

diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index 9f1a93f..bbc3bc6 100644
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -102,8 +102,21 @@ out:
 	preempt_enable();
 }
 
-int perf_output_begin(struct perf_output_handle *handle,
-		      struct perf_event *event, unsigned int size)
+static bool __always_inline
+ring_buffer_has_space(unsigned long head, unsigned long tail,
+		      unsigned long data_size, unsigned int size,
+		      bool backward)
+{
+	if (!backward)
+		return CIRC_SPACE(head, tail, data_size) < size;
+	else
+		return CIRC_SPACE(tail, head, data_size) < size;
+}
+
+static int __always_inline
+__perf_output_begin(struct perf_output_handle *handle,
+		    struct perf_event *event, unsigned int size,
+		    bool backward)
 {
 	struct ring_buffer *rb;
 	unsigned long tail, offset, head;
@@ -146,9 +159,12 @@ int perf_output_begin(struct perf_output_handle *handle,
 	do {
 		tail = READ_ONCE(rb->user_page->data_tail);
 		offset = head = local_read(&rb->head);
-		if (!rb->overwrite &&
-		    unlikely(CIRC_SPACE(head, tail, perf_data_size(rb)) < size))
-			goto fail;
+		if (!rb->overwrite) {
+			if (unlikely(!ring_buffer_has_space(head, tail,
+							    perf_data_size(rb),
+							    size, backward)))
+				goto fail;
+		}
 
 		/*
 		 * The above forms a control dependency barrier separating the
@@ -162,7 +178,10 @@ int perf_output_begin(struct perf_output_handle *handle,
 		 * See perf_output_put_handle().
 		 */
 
-		head += size;
+		if (!backward)
+			head += size;
+		else
+			head -= size;
 	} while (local_cmpxchg(&rb->head, offset, head) != offset);
 
 	/*
@@ -206,6 +225,12 @@ out:
 	return -ENOSPC;
 }
 
+int perf_output_begin(struct perf_output_handle *handle,
+		      struct perf_event *event, unsigned int size)
+{
+	return __perf_output_begin(handle, event, size, false);
+}
+
 unsigned int perf_output_copy(struct perf_output_handle *handle,
 		      const void *buf, unsigned int len)
 {
-- 
1.8.3.4

[toc] | [prev] | [next] | [standalone]

#1312270 — Re: [PATCH 0/6] perf core: Read from overwrite ring buffer

Hi,

On Tue, Jan 19, 2016 at 11:16:44AM +0000, Wang Nan wrote:
> This patchset introduces two methods to support reading from overwrite.
> 
>  1) Tailsize: write the size of an event at the end of it
>  2) Backward writing: write the ring buffer from the end of it to the
>     beginning.

So both of two methods should be used together?

Thanks,
Namhyung


> 
> Patch 1/6 introduces a new ioctl operation to pause and resume ring
> buffer since reading from a overwrite ring buffer is not reliable.
> 
> To reduce overhead as much as possible, force setting overflow_handler
> and create specific function for backward writing and onward writing.
> 
> Cc: He Kuang <hekuang@huawei.com>
> Cc: Alexei Starovoitov <ast@kernel.org>
> Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
> Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
> Cc: Jiri Olsa <jolsa@kernel.org>
> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
> Cc: Namhyung Kim <namhyung@kernel.org>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Zefan Li <lizefan@huawei.com>
> Cc: pi3orama@163.com
> 
> Wang Nan (6):
>   perf core: Introduce new ioctl options to pause and resume ring buffer
>   perf core: Set event's default overflow_handler
>   perf core: Prepare writing into ring buffer from end
>   perf core: Add backwork attribute to perf event
>   perf core: Reduce perf event output overhead by setting overwrite
>     handler
>   perf/core: Put size of a sample at the end of it by
>     PERF_SAMPLE_TAILSIZE
> 
>  include/linux/perf_event.h      |  39 +++++++---
>  include/uapi/linux/perf_event.h |   7 +-
>  kernel/events/core.c            | 155 +++++++++++++++++++++++++++++++---------
>  kernel/events/internal.h        |  11 +++
>  kernel/events/ring_buffer.c     |  65 ++++++++++++++---
>  5 files changed, 223 insertions(+), 54 deletions(-)
> 
> -- 
> 1.8.3.4
> 

[toc] | [prev] | [next] | [standalone]

#1312276 — Re: [PATCH 0/6] perf core: Read from overwrite ring buffer

发自我的 iPhone

> 在 2016年1月19日，下午9:58，Namhyung Kim <namhyung@kernel.org> 写道：
> 
> Hi,
> 
>> On Tue, Jan 19, 2016 at 11:16:44AM +0000, Wang Nan wrote:
>> This patchset introduces two methods to support reading from overwrite.
>> 
>> 1) Tailsize: write the size of an event at the end of it
>> 2) Backward writing: write the ring buffer from the end of it to the
>>    beginning.
> 
> So both of two methods should be used together?
> 

They are separated, we should use only one
of them. I prefer backward writing. But if we
select them both they should work.

Both of them have drawback. Tailsize
method adds 8 bytes to each record
(needs more cycles to write and consume
more space), backward writing may causes
more cache misses since processor doesn't
support backward cache prefetching. Tomorrow I will compare their performance penalty to see which one is better.

Thank you.

> Thanks,
> Namhyung
> 
> 
>> 
>> Patch 1/6 introduces a new ioctl operation to pause and resume ring
>> buffer since reading from a overwrite ring buffer is not reliable.
>> 
>> To reduce overhead as much as possible, force setting overflow_handler
>> and create specific function for backward writing and onward writing.
>> 
>> Cc: He Kuang <hekuang@huawei.com>
>> Cc: Alexei Starovoitov <ast@kernel.org>
>> Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
>> Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
>> Cc: Jiri Olsa <jolsa@kernel.org>
>> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
>> Cc: Namhyung Kim <namhyung@kernel.org>
>> Cc: Peter Zijlstra <peterz@infradead.org>
>> Cc: Zefan Li <lizefan@huawei.com>
>> Cc: pi3orama@163.com
>> 
>> Wang Nan (6):
>>  perf core: Introduce new ioctl options to pause and resume ring buffer
>>  perf core: Set event's default overflow_handler
>>  perf core: Prepare writing into ring buffer from end
>>  perf core: Add backwork attribute to perf event
>>  perf core: Reduce perf event output overhead by setting overwrite
>>    handler
>>  perf/core: Put size of a sample at the end of it by
>>    PERF_SAMPLE_TAILSIZE
>> 
>> include/linux/perf_event.h      |  39 +++++++---
>> include/uapi/linux/perf_event.h |   7 +-
>> kernel/events/core.c            | 155 +++++++++++++++++++++++++++++++---------
>> kernel/events/internal.h        |  11 +++
>> kernel/events/ring_buffer.c     |  65 ++++++++++++++---
>> 5 files changed, 223 insertions(+), 54 deletions(-)
>> 
>> -- 
>> 1.8.3.4
>> 

[toc] | [prev] | [next] | [standalone]

#1312403 — Re: [PATCH 0/6] perf core: Read from overwrite ring buffer

On Tue, Jan 19, 2016 at 11:16:44AM +0000, Wang Nan wrote:
> This patchset introduces two methods to support reading from overwrite.
> 
>  1) Tailsize: write the size of an event at the end of it
>  2) Backward writing: write the ring buffer from the end of it to the
>     beginning.

what happend with your other idea of moving the whole header to the end?
That felt better than either of these options.

[toc] | [prev] | [next] | [standalone]

#1312746 — Re: [PATCH 0/6] perf core: Read from overwrite ring buffer

On 2016/1/20 1:42, Alexei Starovoitov wrote:
> On Tue, Jan 19, 2016 at 11:16:44AM +0000, Wang Nan wrote:
>> This patchset introduces two methods to support reading from overwrite.
>>
>>   1) Tailsize: write the size of an event at the end of it
>>   2) Backward writing: write the ring buffer from the end of it to the
>>      beginning.
> what happend with your other idea of moving the whole header to the end?
> That felt better than either of these options.

I'll try it today. However, putting all of the three together is
not as easy as this patchset.

Thank you.

[toc] | [prev] | [next] | [standalone]

#1312829 — Re: [PATCH 0/6] perf core: Read from overwrite ring buffer

On Wed, Jan 20, 2016 at 09:37:42AM +0800, Wangnan (F) wrote:
> 
> 
> On 2016/1/20 1:42, Alexei Starovoitov wrote:
> >On Tue, Jan 19, 2016 at 11:16:44AM +0000, Wang Nan wrote:
> >>This patchset introduces two methods to support reading from overwrite.
> >>
> >>  1) Tailsize: write the size of an event at the end of it
> >>  2) Backward writing: write the ring buffer from the end of it to the
> >>     beginning.
> >what happend with your other idea of moving the whole header to the end?
> >That felt better than either of these options.
> 
> I'll try it today. However, putting all of the three together is
> not as easy as this patchset.

I'm missing something. Why all three in one set?
Since you have 1 and 2 implemented, benchmark them with absolute
numbers and then implement this last one without any prior baggage
and benchmark it as well. I think it should be the fastest
and the cleanest. We don't need ten different ways to do one thing.

[toc] | [prev] | [next] | [standalone]

#1313963 — Re: [PATCH 0/6] perf core: Read from overwrite ring buffer

On 2016/1/20 10:20, Alexei Starovoitov wrote:
> On Wed, Jan 20, 2016 at 09:37:42AM +0800, Wangnan (F) wrote:
>>
>> On 2016/1/20 1:42, Alexei Starovoitov wrote:
>>> On Tue, Jan 19, 2016 at 11:16:44AM +0000, Wang Nan wrote:
>>>> This patchset introduces two methods to support reading from overwrite.
>>>>
>>>>   1) Tailsize: write the size of an event at the end of it
>>>>   2) Backward writing: write the ring buffer from the end of it to the
>>>>      beginning.
>>> what happend with your other idea of moving the whole header to the end?
>>> That felt better than either of these options.
>> I'll try it today. However, putting all of the three together is
>> not as easy as this patchset.
> I'm missing something. Why all three in one set?

Can't implement all three in one, but implement two of them make
benchmarking simpler :)

Here comes some numbers.

I attach a target program at the end of this mail. It calls
close(-1) for 3000000 times, and use gettimeofday to check
how many us it takes.

Following cases are tested:


  BASE    : ./a.out
  RAWPERF : ./perf record -o /dev/null -e raw_syscalls:* ./a.out
  WRTBKWRD: ./perf record -o /dev/null -e raw_syscalls:* ./a.out
  TAILSIZE: ./perf record --no-has-write-backward -o /dev/null -e 
raw_syscalls:*/overwrite/ ./a.out
  RAWOVWRT: ./perf record --no-has-write-backward --no-has-tailsize -o 
/dev/null -e raw_syscalls:*/overwrite/ ./a.out

With this script:

func() {
         for x in `seq 1 100` ; do $1; done | tee data_$2
}

func ./a.out base
func "./perf record -o /dev/null -e raw_syscalls:* ./a.out" rawperf
func "./perf record -o /dev/null -e raw_syscalls:*/overwrite/ ./a.out" 
wrtbkwrd
func "./perf record -o /dev/null --no-has-write-backward -e 
raw_syscalls:*/overwrite/ ./a.out" tailsize
func "./perf record -o /dev/null --no-has-write-backward 
--no-has-tailsize -o /dev/null -e raw_syscalls:*/overwrite/ ./a.out" 
rawovwrt

Result:

             MEAN           STDVAR
BASE    :  879870.81      11913.13
RAWPERF : 2603854.7      706658.4
WRTBKWRD: 2313301.220      6727.957
TAILSIZE: 2383051.860      5248.061
RAWOVWRT: 2315273.180      5221.025

So it seems backward writing methods is good enough. We don't need to 
consider
tailsize method.

Code for this benchmark can be found from:

https://git.kernel.org/cgit/linux/kernel/git/pi3orama/linux.git/ 
perf/overwrite-benchmark

Thank you.

-------- Test program ----------
#include <unistd.h>
#include <fcntl.h>
#include <sys/time.h>
#include <stdio.h>

int main()
{
         int i;
         struct timeval tv1, tv2;
         long long us1, us2;

         gettimeofday(&tv1, NULL);
         for (i = 0; i < 1000 * 1000 * 3; i++) {
                 close(-1);
         }
         gettimeofday(&tv2, NULL);

         us1 = tv1.tv_sec * 1000000 + tv1.tv_usec;
         us2 = tv2.tv_sec * 1000000 + tv2.tv_usec;
         printf("%ld\n", us2 - us1);

         return 0;
}

[toc] | [prev] | [next] | [standalone]

#1314715 — Re: [PATCH 0/6] perf core: Read from overwrite ring buffer

On 2016/1/21 14:51, Wangnan (F) wrote:
>
>
> On 2016/1/20 10:20, Alexei Starovoitov wrote:
>> On Wed, Jan 20, 2016 at 09:37:42AM +0800, Wangnan (F) wrote:
>>>
>>> On 2016/1/20 1:42, Alexei Starovoitov wrote:
>>>> On Tue, Jan 19, 2016 at 11:16:44AM +0000, Wang Nan wrote:
>>>>> This patchset introduces two methods to support reading from 
>>>>> overwrite.
>>>>>
>>>>>   1) Tailsize: write the size of an event at the end of it
>>>>>   2) Backward writing: write the ring buffer from the end of it to 
>>>>> the
>>>>>      beginning.
>>>> what happend with your other idea of moving the whole header to the 
>>>> end?
>>>> That felt better than either of these options.
>>> I'll try it today. However, putting all of the three together is
>>> not as easy as this patchset.
>> I'm missing something. Why all three in one set?
>
> Can't implement all three in one, but implement two of them make
> benchmarking simpler :)
>
> Here comes some numbers.
>
> I attach a target program at the end of this mail. It calls
> close(-1) for 3000000 times, and use gettimeofday to check
> how many us it takes.
>
> Following cases are tested:
>
>
>  BASE    : ./a.out
>  RAWPERF : ./perf record -o /dev/null -e raw_syscalls:* ./a.out
>  WRTBKWRD: ./perf record -o /dev/null -e raw_syscalls:* ./a.out
>  TAILSIZE: ./perf record --no-has-write-backward -o /dev/null -e 
> raw_syscalls:*/overwrite/ ./a.out
>  RAWOVWRT: ./perf record --no-has-write-backward --no-has-tailsize -o 
> /dev/null -e raw_syscalls:*/overwrite/ ./a.out
>
> With this script:
>
> func() {
>         for x in `seq 1 100` ; do $1; done | tee data_$2
> }
>
> func ./a.out base
> func "./perf record -o /dev/null -e raw_syscalls:* ./a.out" rawperf
> func "./perf record -o /dev/null -e raw_syscalls:*/overwrite/ ./a.out" 
> wrtbkwrd
> func "./perf record -o /dev/null --no-has-write-backward -e 
> raw_syscalls:*/overwrite/ ./a.out" tailsize
> func "./perf record -o /dev/null --no-has-write-backward 
> --no-has-tailsize -o /dev/null -e raw_syscalls:*/overwrite/ ./a.out" 
> rawovwrt
>
> Result:
>
>             MEAN           STDVAR
> BASE    :  879870.81      11913.13
> RAWPERF : 2603854.7      706658.4
> WRTBKWRD: 2313301.220      6727.957
> TAILSIZE: 2383051.860      5248.061
> RAWOVWRT: 2315273.180      5221.025

Add a number: I tested original perf overwrite ring buffer in pure v4.4
on the same machine:

                     MEAN          STDVAR
RAWOVWRT(original): 2323970.45    5103.39

So I think backward writing method doesn't add extra overhead into
fastpath.

I will send this patchset again with several bugs fixed. After that
I'll start working on tail-header if it is still required.

Thank you.

[toc] | [prev] | [next] | [standalone]

#1314727 — Re: [PATCH 0/6] perf core: Read from overwrite ring buffer

On Fri, Jan 22, 2016 at 10:21:19AM +0800, Wangnan (F) wrote:
> 
> 
> On 2016/1/21 14:51, Wangnan (F) wrote:
> >
> >
> >On 2016/1/20 10:20, Alexei Starovoitov wrote:
> >>On Wed, Jan 20, 2016 at 09:37:42AM +0800, Wangnan (F) wrote:
> >>>
> >>>On 2016/1/20 1:42, Alexei Starovoitov wrote:
> >>>>On Tue, Jan 19, 2016 at 11:16:44AM +0000, Wang Nan wrote:
> >>>>>This patchset introduces two methods to support reading from
> >>>>>overwrite.
> >>>>>
> >>>>>  1) Tailsize: write the size of an event at the end of it
> >>>>>  2) Backward writing: write the ring buffer from the end of it to
> >>>>>the
> >>>>>     beginning.
> >>>>what happend with your other idea of moving the whole header to the
> >>>>end?
> >>>>That felt better than either of these options.
> >>>I'll try it today. However, putting all of the three together is
> >>>not as easy as this patchset.
> >>I'm missing something. Why all three in one set?
> >
> >Can't implement all three in one, but implement two of them make
> >benchmarking simpler :)
> >
> >Here comes some numbers.
> >
> >I attach a target program at the end of this mail. It calls
> >close(-1) for 3000000 times, and use gettimeofday to check
> >how many us it takes.
> >
> >Following cases are tested:
> >
> >
> > BASE    : ./a.out
> > RAWPERF : ./perf record -o /dev/null -e raw_syscalls:* ./a.out
> > WRTBKWRD: ./perf record -o /dev/null -e raw_syscalls:* ./a.out
> > TAILSIZE: ./perf record --no-has-write-backward -o /dev/null -e
> >raw_syscalls:*/overwrite/ ./a.out
> > RAWOVWRT: ./perf record --no-has-write-backward --no-has-tailsize -o
> >/dev/null -e raw_syscalls:*/overwrite/ ./a.out
> >
> >With this script:
> >
> >func() {
> >        for x in `seq 1 100` ; do $1; done | tee data_$2
> >}
> >
> >func ./a.out base
> >func "./perf record -o /dev/null -e raw_syscalls:* ./a.out" rawperf
> >func "./perf record -o /dev/null -e raw_syscalls:*/overwrite/ ./a.out"
> >wrtbkwrd
> >func "./perf record -o /dev/null --no-has-write-backward -e
> >raw_syscalls:*/overwrite/ ./a.out" tailsize
> >func "./perf record -o /dev/null --no-has-write-backward --no-has-tailsize
> >-o /dev/null -e raw_syscalls:*/overwrite/ ./a.out" rawovwrt
> >
> >Result:
> >
> >            MEAN           STDVAR
> >BASE    :  879870.81      11913.13
> >RAWPERF : 2603854.7      706658.4
> >WRTBKWRD: 2313301.220      6727.957
> >TAILSIZE: 2383051.860      5248.061
> >RAWOVWRT: 2315273.180      5221.025
> 
> Add a number: I tested original perf overwrite ring buffer in pure v4.4
> on the same machine:
> 
>                     MEAN          STDVAR
> RAWOVWRT(original): 2323970.45    5103.39
> 
> So I think backward writing method doesn't add extra overhead into
> fastpath.
> 
> I will send this patchset again with several bugs fixed. After that
> I'll start working on tail-header if it is still required.

interesting.
did I read the numbers correctly that 'write backwards' method
is actually the fastest? even faster than no-overwrite?
nice. I guess it makes snese that overwrite is faster.
I guess than moving the header to the end will have the same
performance in this benchmark, since RAWOVWRT is the same as well.

[toc] | [prev] | [next] | [standalone]

#1314748 — Re: [PATCH 0/6] perf core: Read from overwrite ring buffer

On 2016/1/22 11:21, Alexei Starovoitov wrote:
> On Fri, Jan 22, 2016 at 10:21:19AM +0800, Wangnan (F) wrote:
>>
>> On 2016/1/21 14:51, Wangnan (F) wrote:
>>>
>>> On 2016/1/20 10:20, Alexei Starovoitov wrote:
>>>> On Wed, Jan 20, 2016 at 09:37:42AM +0800, Wangnan (F) wrote:
>>>>> On 2016/1/20 1:42, Alexei Starovoitov wrote:
>>>>>> On Tue, Jan 19, 2016 at 11:16:44AM +0000, Wang Nan wrote:
>>>>>>> This patchset introduces two methods to support reading from
>>>>>>> overwrite.
>>>>>>>
>>>>>>>   1) Tailsize: write the size of an event at the end of it
>>>>>>>   2) Backward writing: write the ring buffer from the end of it to
>>>>>>> the
>>>>>>>      beginning.
>>>>>> what happend with your other idea of moving the whole header to the
>>>>>> end?
>>>>>> That felt better than either of these options.
>>>>> I'll try it today. However, putting all of the three together is
>>>>> not as easy as this patchset.
>>>> I'm missing something. Why all three in one set?
>>> Can't implement all three in one, but implement two of them make
>>> benchmarking simpler :)
>>>
>>> Here comes some numbers.
>>>
>>> I attach a target program at the end of this mail. It calls
>>> close(-1) for 3000000 times, and use gettimeofday to check
>>> how many us it takes.
>>>
>>> Following cases are tested:
>>>
>>>
>>> BASE    : ./a.out
>>> RAWPERF : ./perf record -o /dev/null -e raw_syscalls:* ./a.out
>>> WRTBKWRD: ./perf record -o /dev/null -e raw_syscalls:* ./a.out
>>> TAILSIZE: ./perf record --no-has-write-backward -o /dev/null -e
>>> raw_syscalls:*/overwrite/ ./a.out
>>> RAWOVWRT: ./perf record --no-has-write-backward --no-has-tailsize -o
>>> /dev/null -e raw_syscalls:*/overwrite/ ./a.out
>>>
>>> With this script:
>>>
>>> func() {
>>>         for x in `seq 1 100` ; do $1; done | tee data_$2
>>> }
>>>
>>> func ./a.out base
>>> func "./perf record -o /dev/null -e raw_syscalls:* ./a.out" rawperf
>>> func "./perf record -o /dev/null -e raw_syscalls:*/overwrite/ ./a.out"
>>> wrtbkwrd
>>> func "./perf record -o /dev/null --no-has-write-backward -e
>>> raw_syscalls:*/overwrite/ ./a.out" tailsize
>>> func "./perf record -o /dev/null --no-has-write-backward --no-has-tailsize
>>> -o /dev/null -e raw_syscalls:*/overwrite/ ./a.out" rawovwrt
>>>
>>> Result:
>>>
>>>             MEAN           STDVAR
>>> BASE    :  879870.81      11913.13
>>> RAWPERF : 2603854.7      706658.4
>>> WRTBKWRD: 2313301.220      6727.957
>>> TAILSIZE: 2383051.860      5248.061
>>> RAWOVWRT: 2315273.180      5221.025
>> Add a number: I tested original perf overwrite ring buffer in pure v4.4
>> on the same machine:
>>
>>                      MEAN          STDVAR
>> RAWOVWRT(original): 2323970.45    5103.39
>>
>> So I think backward writing method doesn't add extra overhead into
>> fastpath.
>>
>> I will send this patchset again with several bugs fixed. After that
>> I'll start working on tail-header if it is still required.
> interesting.
> did I read the numbers correctly that 'write backwards' method
> is actually the fastest? even faster than no-overwrite?

Yes. But notice STDVAR, we can't say 'WRTBKWRD' outperform 'RAWOVWRT'. 
However,
at least 'WRTBKWRD' should be as fast as 'RAWOVWRT'.

> nice. I guess it makes snese that overwrite is faster.

In no-overwrite case perf itself wakes up many times to collect data,
I guess it is the source of high stdvar.

> I guess than moving the header to the end will have the same
> performance in this benchmark, since RAWOVWRT is the same as well.
>
Yes.

Do you want to test it by yourself? The code is ready.

Thank you.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web