[PATCH] perf_events: fix ordering bug in perf_output_sample()

[PATCH] perf_events: fix ordering bug in perf_output_sample()

[Stephane Eranian]

	In order to parse a sample correctly based on the information
	requested via sample_type, the kernel needs to save each component
	in a known order. There is no type value saved with each component.
	The current convention is that each component is saved according to
	the order in enum perf_event_sample_format. But perf_output_sample()
	was not completely following this convention, thereby making samples
	impossible to parse without internal kernel knowledge.

	This patch puts things in the right order.

	Signed-off-by: Stephane Eranian <eranian@google.com>

--
 perf_event.c |   24 ++++++++++++------------
 1 file changed, 12 insertions(+), 12 deletions(-)

diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index 455393e..b35df1d 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -3167,18 +3167,6 @@ void perf_output_sample(struct perf_output_handle *handle,
 	if (sample_type & PERF_SAMPLE_ADDR)
 		perf_output_put(handle, data->addr);
 
-	if (sample_type & PERF_SAMPLE_ID)
-		perf_output_put(handle, data->id);
-
-	if (sample_type & PERF_SAMPLE_STREAM_ID)
-		perf_output_put(handle, data->stream_id);
-
-	if (sample_type & PERF_SAMPLE_CPU)
-		perf_output_put(handle, data->cpu_entry);
-
-	if (sample_type & PERF_SAMPLE_PERIOD)
-		perf_output_put(handle, data->period);
-
 	if (sample_type & PERF_SAMPLE_READ)
 		perf_output_read(handle, event);
 
@@ -3198,6 +3186,18 @@ void perf_output_sample(struct perf_output_handle *handle,
 		}
 	}
 
+	if (sample_type & PERF_SAMPLE_ID)
+		perf_output_put(handle, data->id);
+
+	if (sample_type & PERF_SAMPLE_CPU)
+		perf_output_put(handle, data->cpu_entry);
+
+	if (sample_type & PERF_SAMPLE_STREAM_ID)
+		perf_output_put(handle, data->stream_id);
+
+	if (sample_type & PERF_SAMPLE_PERIOD)
+		perf_output_put(handle, data->period);
+
 	if (sample_type & PERF_SAMPLE_RAW) {
 		if (data->raw) {
 			perf_output_put(handle, data->raw->size);

Re: [PATCH] perf_events: fix ordering bug in perf_output_sample()

[Peter Zijlstra]

On Thu, 2010-03-18 at 14:42 +0200, Stephane Eranian wrote:
> 	In order to parse a sample correctly based on the information
> 	requested via sample_type, the kernel needs to save each component
> 	in a known order. There is no type value saved with each component.
> 	The current convention is that each component is saved according to
> 	the order in enum perf_event_sample_format. But perf_output_sample()
> 	was not completely following this convention, thereby making samples
> 	impossible to parse without internal kernel knowledge.
> 
> 	This patch puts things in the right order.

NAK, not so actually, its in the order specified in the
PERF_RECORD_SAMPLE comment.

Re: [PATCH] perf_events: fix ordering bug in perf_output_sample()

[Stephane Eranian]

On Thu, Mar 18, 2010 at 7:33 PM, Peter Zijlstra <peterz@infradead.org> wrote:
> On Thu, 2010-03-18 at 14:42 +0200, Stephane Eranian wrote:
>>       In order to parse a sample correctly based on the information
>>       requested via sample_type, the kernel needs to save each component
>>       in a known order. There is no type value saved with each component.
>>       The current convention is that each component is saved according to
>>       the order in enum perf_event_sample_format. But perf_output_sample()
>>       was not completely following this convention, thereby making samples
>>       impossible to parse without internal kernel knowledge.
>>
>>       This patch puts things in the right order.
>
> NAK, not so actually, its in the order specified in the
> PERF_RECORD_SAMPLE comment.
>
And why is that order different than the one in the enum?

Re: [PATCH] perf_events: fix ordering bug in perf_output_sample()

[Peter Zijlstra]

On Thu, 2010-03-18 at 22:29 +0100, Stephane Eranian wrote:
> On Thu, Mar 18, 2010 at 7:33 PM, Peter Zijlstra <peterz@infradead.org> wrote:
> > On Thu, 2010-03-18 at 14:42 +0200, Stephane Eranian wrote:
> >>       In order to parse a sample correctly based on the information
> >>       requested via sample_type, the kernel needs to save each component
> >>       in a known order. There is no type value saved with each component.
> >>       The current convention is that each component is saved according to
> >>       the order in enum perf_event_sample_format. But perf_output_sample()
> >>       was not completely following this convention, thereby making samples
> >>       impossible to parse without internal kernel knowledge.
> >>
> >>       This patch puts things in the right order.
> >
> > NAK, not so actually, its in the order specified in the
> > PERF_RECORD_SAMPLE comment.
> >
> And why is that order different than the one in the enum?

Because I tried to keep the simple elements before the complex ones, but
in any case its too late to change any ordering there now, as its ABI.