diff mbox series

[V3,14/14] coresight: etm-perf: Add support for trace buffer format

Message ID 1611737738-1493-15-git-send-email-anshuman.khandual@arm.com (mailing list archive)
State New, archived
Headers show
Series arm64: coresight: Enable ETE and TRBE | expand

Commit Message

Anshuman Khandual Jan. 27, 2021, 8:55 a.m. UTC 
From: Suzuki K Poulose <suzuki.poulose@arm.com>

CoreSight PMU supports aux-buffer for the ETM tracing. The trace
generated by the ETM (associated with individual CPUs, like Intel PT)
is captured by a separate IP (CoreSight TMC-ETR/ETF until now).

The TMC-ETR applies formatting of the raw ETM trace data, as it
can collect traces from multiple ETMs, with the TraceID to indicate
the source of a given trace packet.

Arm Trace Buffer Extension is new "sink" IP, attached to individual
CPUs and thus do not provide additional formatting, like TMC-ETR.

Additionally, a system could have both TRBE *and* TMC-ETR for
the trace collection. e.g, TMC-ETR could be used as a single
trace buffer to collect data from multiple ETMs to correlate
the traces from different CPUs. It is possible to have a
perf session where some events end up collecting the trace
in TMC-ETR while the others in TRBE. Thus we need a way
to identify the type of the trace for each AUX record.

Define the trace formats exported by the CoreSight PMU.
We don't define the flags following the "ETM" as this
information is available to the user when issuing
the session. What is missing is the additional
formatting applied by the "sink" which is decided
at the runtime and the user may not have a control on.

So we define :
 - CORESIGHT format (indicates the Frame format)
 - RAW format (indicates the format of the source)

The default value is CORESIGHT format for all the records
(i,e == 0). Add the RAW format for the TRBE sink driver.

Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Leach <mike.leach@linaro.org>
Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
Cc: Leo Yan <leo.yan@linaro.org>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
---
 drivers/hwtracing/coresight/coresight-trbe.c | 2 ++
 include/uapi/linux/perf_event.h              | 4 ++++
 2 files changed, 6 insertions(+)

Comments

Peter Zijlstra Jan. 27, 2021, 12:54 p.m. UTC | #1 
On Wed, Jan 27, 2021 at 02:25:38PM +0530, Anshuman Khandual wrote:

> It is possible to have a
> perf session where some events end up collecting the trace
> in TMC-ETR while the others in TRBE. Thus we need a way
> to identify the type of the trace for each AUX record.

> diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
> index 9a5ca45..169e6b3 100644
> --- a/include/uapi/linux/perf_event.h
> +++ b/include/uapi/linux/perf_event.h
> @@ -1111,6 +1111,10 @@ enum perf_callchain_context {
>  #define PERF_AUX_FLAG_COLLISION			0x08	/* sample collided with another */
>  #define PERF_AUX_FLAG_PMU_FORMAT_TYPE_MASK	0xff00	/* PMU specific trace format type */
>  
> +/* CoreSight PMU AUX buffer formats */
> +#define PERF_AUX_FLAG_CORESIGHT_FORMAT_CORESIGHT	0x0000 /* Default for backward compatibility */
> +#define PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW		0x0100 /* Raw format of the source */

Would CORESIGHT_FORMAT_ETR / CORESIGHT_FORMAT_TRBE be better names?
Al Grant Jan. 27, 2021, 1 p.m. UTC | #2 
> > +/* CoreSight PMU AUX buffer formats */
> > +#define PERF_AUX_FLAG_CORESIGHT_FORMAT_CORESIGHT	0x0000 /*
> Default for backward compatibility */
> > +#define PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW		0x0100 /*
> Raw format of the source */
> 
> Would CORESIGHT_FORMAT_ETR / CORESIGHT_FORMAT_TRBE be better
> names?

Unformatted (raw) streams could be used any time you had a writer dedicated
to a single trace source. So in a situation where you had one ETR per CPU,
it would be appropriate to use an unformatted stream. A TRBE is always
dedicated to a single CPU, but potentially you (i.e. when designing the system)
can do this with any type of trace sink. So the raw/formatted distinction is
really about whether you are combining multiple streams in one buffer or not,
rather than the type of block that is writing into the buffer.

Al
Suzuki K Poulose Jan. 27, 2021, 2:12 p.m. UTC | #3 
Hi Peter

On 1/27/21 12:54 PM, Peter Zijlstra wrote:
> On Wed, Jan 27, 2021 at 02:25:38PM +0530, Anshuman Khandual wrote:
> 
>> It is possible to have a
>> perf session where some events end up collecting the trace
>> in TMC-ETR while the others in TRBE. Thus we need a way
>> to identify the type of the trace for each AUX record.
> 
>> diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
>> index 9a5ca45..169e6b3 100644
>> --- a/include/uapi/linux/perf_event.h
>> +++ b/include/uapi/linux/perf_event.h
>> @@ -1111,6 +1111,10 @@ enum perf_callchain_context {
>>   #define PERF_AUX_FLAG_COLLISION			0x08	/* sample collided with another */
>>   #define PERF_AUX_FLAG_PMU_FORMAT_TYPE_MASK	0xff00	/* PMU specific trace format type */
>>   
>> +/* CoreSight PMU AUX buffer formats */
>> +#define PERF_AUX_FLAG_CORESIGHT_FORMAT_CORESIGHT	0x0000 /* Default for backward compatibility */
>> +#define PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW		0x0100 /* Raw format of the source */
> 
> Would CORESIGHT_FORMAT_ETR / CORESIGHT_FORMAT_TRBE be better names?
> 

Not really. The format used by the ETR is CORESIGHT frame formatting. And
there are other sinks that uses the formatting. e.g, TMC-ETB (a sink with
internal memory), TMC-ETF (trace fifo with internal memory). So it is really
not tied to ETR.

As for TRBE, it simply pumps the data thrown at it to the memory. As such
calling it TRBE format would be confusing as the format of the buffer is
really Raw trace thrown at it. This can be inferred from the ETM/ETE
looking at the ID registers, which the userspace perf already captures
in the perf.data. So the decoder perf can look at the perf.data and the
AUX records to interpret the buffer correctly.

Suzuki
Mike Leach Feb. 16, 2021, 11:01 a.m. UTC | #4 
On Wed, 27 Jan 2021 at 08:56, Anshuman Khandual
<anshuman.khandual@arm.com> wrote:
>
> From: Suzuki K Poulose <suzuki.poulose@arm.com>
>
> CoreSight PMU supports aux-buffer for the ETM tracing. The trace
> generated by the ETM (associated with individual CPUs, like Intel PT)
> is captured by a separate IP (CoreSight TMC-ETR/ETF until now).
>
> The TMC-ETR applies formatting of the raw ETM trace data, as it
> can collect traces from multiple ETMs, with the TraceID to indicate
> the source of a given trace packet.
>
> Arm Trace Buffer Extension is new "sink" IP, attached to individual
> CPUs and thus do not provide additional formatting, like TMC-ETR.
>
> Additionally, a system could have both TRBE *and* TMC-ETR for
> the trace collection. e.g, TMC-ETR could be used as a single
> trace buffer to collect data from multiple ETMs to correlate
> the traces from different CPUs. It is possible to have a
> perf session where some events end up collecting the trace
> in TMC-ETR while the others in TRBE. Thus we need a way
> to identify the type of the trace for each AUX record.
>
> Define the trace formats exported by the CoreSight PMU.
> We don't define the flags following the "ETM" as this
> information is available to the user when issuing
> the session. What is missing is the additional
> formatting applied by the "sink" which is decided
> at the runtime and the user may not have a control on.
>
> So we define :
>  - CORESIGHT format (indicates the Frame format)
>  - RAW format (indicates the format of the source)
>
> The default value is CORESIGHT format for all the records
> (i,e == 0). Add the RAW format for the TRBE sink driver.
>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Mike Leach <mike.leach@linaro.org>
> Cc: Mathieu Poirier <mathieu.poirier@linaro.org>
> Cc: Leo Yan <leo.yan@linaro.org>
> Cc: Anshuman Khandual <anshuman.khandual@arm.com>
> Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
> Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
> ---
>  drivers/hwtracing/coresight/coresight-trbe.c | 2 ++
>  include/uapi/linux/perf_event.h              | 4 ++++
>  2 files changed, 6 insertions(+)
>
> diff --git a/drivers/hwtracing/coresight/coresight-trbe.c b/drivers/hwtracing/coresight/coresight-trbe.c
> index 1464d8b..7c0e691 100644
> --- a/drivers/hwtracing/coresight/coresight-trbe.c
> +++ b/drivers/hwtracing/coresight/coresight-trbe.c
> @@ -511,6 +511,7 @@ static unsigned long arm_trbe_update_buffer(struct coresight_device *csdev,
>         if (cpudata->mode != CS_MODE_PERF)
>                 return -EINVAL;
>
> +       perf_aux_output_flag(handle, PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW);
>         /*
>          * If the TRBE was disabled due to lack of space in the AUX buffer or a
>          * spurious fault, the driver leaves it disabled, truncating the buffer.
> @@ -606,6 +607,7 @@ static void trbe_handle_overflow(struct perf_output_handle *handle)
>         size = offset - PERF_IDX2OFF(handle->head, buf);
>         if (buf->snapshot)
>                 handle->head = offset;
> +       perf_aux_output_flag(handle, PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW);
>         perf_aux_output_end(handle, size);
>
>         event_data = perf_aux_output_begin(handle, event);
> diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
> index 9a5ca45..169e6b3 100644
> --- a/include/uapi/linux/perf_event.h
> +++ b/include/uapi/linux/perf_event.h
> @@ -1111,6 +1111,10 @@ enum perf_callchain_context {
>  #define PERF_AUX_FLAG_COLLISION                        0x08    /* sample collided with another */
>  #define PERF_AUX_FLAG_PMU_FORMAT_TYPE_MASK     0xff00  /* PMU specific trace format type */
>
> +/* CoreSight PMU AUX buffer formats */
> +#define PERF_AUX_FLAG_CORESIGHT_FORMAT_CORESIGHT       0x0000 /* Default for backward compatibility */
> +#define PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW             0x0100 /* Raw format of the source */
> +
>  #define PERF_FLAG_FD_NO_GROUP          (1UL << 0)
>  #define PERF_FLAG_FD_OUTPUT            (1UL << 1)
>  #define PERF_FLAG_PID_CGROUP           (1UL << 2) /* pid=cgroup id, per-cpu mode only */
> --
> 2.7.4
>

Reviewed-by: Mike Leach <mike.leach@linaro.org>
Anshuman Khandual Feb. 18, 2021, 3:05 a.m. UTC | #5 
On 1/27/21 6:30 PM, Al Grant wrote:
>>> +/* CoreSight PMU AUX buffer formats */
>>> +#define PERF_AUX_FLAG_CORESIGHT_FORMAT_CORESIGHT0x0000 /*
>> Default for backward compatibility */
>>> +#define PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW0x0100 /*
>> Raw format of the source */
>>
>> Would CORESIGHT_FORMAT_ETR / CORESIGHT_FORMAT_TRBE be better
>> names?
> 
> Unformatted (raw) streams could be used any time you had a writer dedicated
> to a single trace source. So in a situation where you had one ETR per CPU,
> it would be appropriate to use an unformatted stream. A TRBE is always
> dedicated to a single CPU, but potentially you (i.e. when designing the system)
> can do this with any type of trace sink. So the raw/formatted distinction is
> really about whether you are combining multiple streams in one buffer or not,
> rather than the type of block that is writing into the buffer.
> 
> Al
> 

Okay, will stick with the proposed format names here

i.e 

PERF_AUX_FLAG_CORESIGHT_FORMAT_CORESIGHT
PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW
diff mbox series

Patch

diff --git a/drivers/hwtracing/coresight/coresight-trbe.c b/drivers/hwtracing/coresight/coresight-trbe.c
index 1464d8b..7c0e691 100644
--- a/drivers/hwtracing/coresight/coresight-trbe.c
+++ b/drivers/hwtracing/coresight/coresight-trbe.c
@@ -511,6 +511,7 @@  static unsigned long arm_trbe_update_buffer(struct coresight_device *csdev,
 	if (cpudata->mode != CS_MODE_PERF)
 		return -EINVAL;
 
+	perf_aux_output_flag(handle, PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW);
 	/*
 	 * If the TRBE was disabled due to lack of space in the AUX buffer or a
 	 * spurious fault, the driver leaves it disabled, truncating the buffer.
@@ -606,6 +607,7 @@  static void trbe_handle_overflow(struct perf_output_handle *handle)
 	size = offset - PERF_IDX2OFF(handle->head, buf);
 	if (buf->snapshot)
 		handle->head = offset;
+	perf_aux_output_flag(handle, PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW);
 	perf_aux_output_end(handle, size);
 
 	event_data = perf_aux_output_begin(handle, event);
diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index 9a5ca45..169e6b3 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -1111,6 +1111,10 @@  enum perf_callchain_context {
 #define PERF_AUX_FLAG_COLLISION			0x08	/* sample collided with another */
 #define PERF_AUX_FLAG_PMU_FORMAT_TYPE_MASK	0xff00	/* PMU specific trace format type */
 
+/* CoreSight PMU AUX buffer formats */
+#define PERF_AUX_FLAG_CORESIGHT_FORMAT_CORESIGHT	0x0000 /* Default for backward compatibility */
+#define PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW		0x0100 /* Raw format of the source */
+
 #define PERF_FLAG_FD_NO_GROUP		(1UL << 0)
 #define PERF_FLAG_FD_OUTPUT		(1UL << 1)
 #define PERF_FLAG_PID_CGROUP		(1UL << 2) /* pid=cgroup id, per-cpu mode only */