[PATCH v1 0/2] perf cs-etm: Improve timestamp tracing

[PATCH v1 0/2] perf cs-etm: Improve timestamp tracing

[Leo Yan]

This patch series is to improve timestamp handling in per-thread mode.

The current code doesn't validate timestamp and always return success for
per-thread mode, for a sane implementation, the first patch is to allow
validation timestamp tracing in per-thread mode.

The second patch is to respect timestamp option "--timestamp" or "-T",
when users set this option, the tool will automatically enable hardware
timestamp tracing in Arm CoreSight.


Leo Yan (2):
  perf cs-etm: Validate timestamp tracing in per-thread mode
  perf cs-etm: Respect timestamp option

 tools/perf/arch/arm/util/cs-etm.c | 22 ++++++++++++++++++++--
 1 file changed, 20 insertions(+), 2 deletions(-)

-- 
2.34.1

[PATCH v1 1/2] perf cs-etm: Validate timestamp tracing in per-thread mode

[Leo Yan]

So far, it's impossible to validate timestamp trace in Arm CoreSight when
the perf is in the per-thread mode.  E.g. for the command:

  perf record -e cs_etm/timestamp/ --per-thread -- ls

The command enables config 'timestamp' for 'cs_etm' event in the
per-thread mode.  In this case, the function cs_etm_validate_config()
directly bails out and skips validation.

Given profiled process can be scheduled on any CPUs in the per-thread
mode, this patch validates timestamp tracing for all CPUs when detect
the CPU map is empty.

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 tools/perf/arch/arm/util/cs-etm.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/tools/perf/arch/arm/util/cs-etm.c b/tools/perf/arch/arm/util/cs-etm.c
index b8d6a953fd74..cf9ef9ba800b 100644
--- a/tools/perf/arch/arm/util/cs-etm.c
+++ b/tools/perf/arch/arm/util/cs-etm.c
@@ -205,8 +205,17 @@ static int cs_etm_validate_config(struct auxtrace_record *itr,
 	for (i = 0; i < cpu__max_cpu().cpu; i++) {
 		struct perf_cpu cpu = { .cpu = i, };
 
-		if (!perf_cpu_map__has(event_cpus, cpu) ||
-		    !perf_cpu_map__has(online_cpus, cpu))
+		/*
+		 * In per-cpu case, do the validation for CPUs to work with.
+		 * In per-thread case, the CPU map is empty.  Since the traced
+		 * program can run on any CPUs in this case, thus don't skip
+		 * validation.
+		 */
+		if (!perf_cpu_map__empty(event_cpus) &&
+		    !perf_cpu_map__has(event_cpus, cpu))
+			continue;
+
+		if (!perf_cpu_map__has(online_cpus, cpu))
 			continue;
 
 		err = cs_etm_validate_context_id(itr, evsel, i);
-- 
2.34.1

[PATCH v1 2/2] perf cs-etm: Respect timestamp option

[Leo Yan]

When users pass the option '--timestamp' or '-T' in the record command,
all events will set the PERF_SAMPLE_TIME bit in the attribution.  In
this case, the AUX event will record the kernel timestamp, but it
doesn't mean Arm CoreSight enables timestamp packets in its hardware
tracing.

If the option '--timestamp' or '-T' is set, this patch always enables
Arm CoreSight timestamp, as a result, the bit 28 in event's config is to
be set.

Before:

  # perf record -e cs_etm// --per-thread --timestamp -- ls
  # perf script --header-only
  ...
  # event : name = cs_etm//, , id = { 69 }, type = 12, size = 136,
  config = 0, { sample_period, sample_freq } = 1,
  sample_type = IP|TID|TIME|CPU|IDENTIFIER, read_format = ID|LOST,
  disabled = 1, enable_on_exec = 1, sample_id_all = 1, exclude_guest = 1
  ...

After:

  # perf record -e cs_etm// --per-thread --timestamp -- ls
  # perf script --header-only
  ...
  # event : name = cs_etm//, , id = { 49 }, type = 12, size = 136,
  config = 0x10000000, { sample_period, sample_freq } = 1,
  sample_type = IP|TID|TIME|CPU|IDENTIFIER, read_format = ID|LOST,
  disabled = 1, enable_on_exec = 1, sample_id_all = 1, exclude_guest = 1
  ...

Signed-off-by: Leo Yan <leo.yan@linaro.org>
---
 tools/perf/arch/arm/util/cs-etm.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/tools/perf/arch/arm/util/cs-etm.c b/tools/perf/arch/arm/util/cs-etm.c
index cf9ef9ba800b..58c506e9788d 100644
--- a/tools/perf/arch/arm/util/cs-etm.c
+++ b/tools/perf/arch/arm/util/cs-etm.c
@@ -442,6 +442,15 @@ static int cs_etm_recording_options(struct auxtrace_record *itr,
 					   "contextid", 1);
 	}
 
+	/*
+	 * When the option '--timestamp' or '-T' is enabled, the PERF_SAMPLE_TIME
+	 * bit is set for all events.  In this case, always enable Arm CoreSight
+	 * timestamp tracing.
+	 */
+	if (opts->sample_time_set)
+		evsel__set_config_if_unset(cs_etm_pmu, cs_etm_evsel,
+					   "timestamp", 1);
+
 	/* Add dummy event to keep tracking */
 	err = parse_event(evlist, "dummy:u");
 	if (err)
-- 
2.34.1

Re: [PATCH v1 1/2] perf cs-etm: Validate timestamp tracing in per-thread mode

[James Clark]

On 27/08/2023 14:35, Leo Yan wrote:
> So far, it's impossible to validate timestamp trace in Arm CoreSight when
> the perf is in the per-thread mode.  E.g. for the command:
> 
>   perf record -e cs_etm/timestamp/ --per-thread -- ls
> 
> The command enables config 'timestamp' for 'cs_etm' event in the
> per-thread mode.  In this case, the function cs_etm_validate_config()
> directly bails out and skips validation.
> 
> Given profiled process can be scheduled on any CPUs in the per-thread
> mode, this patch validates timestamp tracing for all CPUs when detect
> the CPU map is empty.

There is an edge case where the profiled process is known by the user to
be pinned to a specific CPU, rather than possibly running on all CPUs,
so this isn't always true.

But I think that can be worked around by changing it to a per-cpu
session to get around the new error. Given that this validation was only
supposed to be best effort information and not get in the way you could
say to not make it more restrictive.

But it's quite a small edge case so either way:

Reviewed-by: James Clark <james.clark@arm.com>

> 
> Signed-off-by: Leo Yan <leo.yan@linaro.org>
> ---
>  tools/perf/arch/arm/util/cs-etm.c | 13 +++++++++++--
>  1 file changed, 11 insertions(+), 2 deletions(-)
> 
> diff --git a/tools/perf/arch/arm/util/cs-etm.c b/tools/perf/arch/arm/util/cs-etm.c
> index b8d6a953fd74..cf9ef9ba800b 100644
> --- a/tools/perf/arch/arm/util/cs-etm.c
> +++ b/tools/perf/arch/arm/util/cs-etm.c
> @@ -205,8 +205,17 @@ static int cs_etm_validate_config(struct auxtrace_record *itr,
>  	for (i = 0; i < cpu__max_cpu().cpu; i++) {
>  		struct perf_cpu cpu = { .cpu = i, };
>  
> -		if (!perf_cpu_map__has(event_cpus, cpu) ||
> -		    !perf_cpu_map__has(online_cpus, cpu))
> +		/*
> +		 * In per-cpu case, do the validation for CPUs to work with.
> +		 * In per-thread case, the CPU map is empty.  Since the traced
> +		 * program can run on any CPUs in this case, thus don't skip
> +		 * validation.
> +		 */
> +		if (!perf_cpu_map__empty(event_cpus) &&
> +		    !perf_cpu_map__has(event_cpus, cpu))
> +			continue;
> +
> +		if (!perf_cpu_map__has(online_cpus, cpu))
>  			continue;
>  
>  		err = cs_etm_validate_context_id(itr, evsel, i);

Re: [PATCH v1 2/2] perf cs-etm: Respect timestamp option

[James Clark]

On 27/08/2023 14:35, Leo Yan wrote:
> When users pass the option '--timestamp' or '-T' in the record command,
> all events will set the PERF_SAMPLE_TIME bit in the attribution.  In
> this case, the AUX event will record the kernel timestamp, but it
> doesn't mean Arm CoreSight enables timestamp packets in its hardware
> tracing.
> 
> If the option '--timestamp' or '-T' is set, this patch always enables
> Arm CoreSight timestamp, as a result, the bit 28 in event's config is to
> be set.
> 
> Before:
> 
>   # perf record -e cs_etm// --per-thread --timestamp -- ls
>   # perf script --header-only
>   ...
>   # event : name = cs_etm//, , id = { 69 }, type = 12, size = 136,
>   config = 0, { sample_period, sample_freq } = 1,
>   sample_type = IP|TID|TIME|CPU|IDENTIFIER, read_format = ID|LOST,
>   disabled = 1, enable_on_exec = 1, sample_id_all = 1, exclude_guest = 1
>   ...
> 
> After:
> 
>   # perf record -e cs_etm// --per-thread --timestamp -- ls
>   # perf script --header-only
>   ...
>   # event : name = cs_etm//, , id = { 49 }, type = 12, size = 136,
>   config = 0x10000000, { sample_period, sample_freq } = 1,
>   sample_type = IP|TID|TIME|CPU|IDENTIFIER, read_format = ID|LOST,
>   disabled = 1, enable_on_exec = 1, sample_id_all = 1, exclude_guest = 1
>   ...
> 
> Signed-off-by: Leo Yan <leo.yan@linaro.org>

Reviewed-by: James Clark <james.clark@arm.com>

> ---
>  tools/perf/arch/arm/util/cs-etm.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
> 
> diff --git a/tools/perf/arch/arm/util/cs-etm.c b/tools/perf/arch/arm/util/cs-etm.c
> index cf9ef9ba800b..58c506e9788d 100644
> --- a/tools/perf/arch/arm/util/cs-etm.c
> +++ b/tools/perf/arch/arm/util/cs-etm.c
> @@ -442,6 +442,15 @@ static int cs_etm_recording_options(struct auxtrace_record *itr,
>  					   "contextid", 1);
>  	}
>  
> +	/*
> +	 * When the option '--timestamp' or '-T' is enabled, the PERF_SAMPLE_TIME
> +	 * bit is set for all events.  In this case, always enable Arm CoreSight
> +	 * timestamp tracing.
> +	 */
> +	if (opts->sample_time_set)
> +		evsel__set_config_if_unset(cs_etm_pmu, cs_etm_evsel,
> +					   "timestamp", 1);
> +
>  	/* Add dummy event to keep tracking */
>  	err = parse_event(evlist, "dummy:u");
>  	if (err)

Re: [PATCH v1 1/2] perf cs-etm: Validate timestamp tracing in per-thread mode

[Leo Yan]

On Mon, Sep 04, 2023 at 04:23:43PM +0100, James Clark wrote:
> 
> On 27/08/2023 14:35, Leo Yan wrote:
> > So far, it's impossible to validate timestamp trace in Arm CoreSight when
> > the perf is in the per-thread mode.  E.g. for the command:
> > 
> >   perf record -e cs_etm/timestamp/ --per-thread -- ls
> > 
> > The command enables config 'timestamp' for 'cs_etm' event in the
> > per-thread mode.  In this case, the function cs_etm_validate_config()
> > directly bails out and skips validation.
> > 
> > Given profiled process can be scheduled on any CPUs in the per-thread
> > mode, this patch validates timestamp tracing for all CPUs when detect
> > the CPU map is empty.
> 
> There is an edge case where the profiled process is known by the user to
> be pinned to a specific CPU, rather than possibly running on all CPUs,
> so this isn't always true.

Good point.

However, when a process is pinned to specific CPUs, we still can
dynamically change the scheduling affinity to any other CPUs by using
taskset command or calling sched_setaffinity().  From a perf session's
pespective, it is sane to validate timestamp tracing for all online
CPUs for per-thread mode.

> But I think that can be worked around by changing it to a per-cpu
> session to get around the new error. Given that this validation was only
> supposed to be best effort information and not get in the way you could
> say to not make it more restrictive.
> 
> But it's quite a small edge case so either way:
> 
> Reviewed-by: James Clark <james.clark@arm.com>

Thanks for review!

Leo