[PATCH v3 0/4] perf: cs-etm: HW_ID improvements

[PATCH v3 0/4] perf: cs-etm: HW_ID improvements

[James Clark]

The driver might not always be able to write HW_IDs, but we don't need
them for unformatted mode anyway, so fix that in commit ("perf cs-etm:
Synthesize missing HW_ID mappings for raw trace").

Include some other minor improvements encountered on the way to appease
Sashiko, like properly freeing queues and ("perf: cs-etm: Respect
--no-itrace option") makes debugging broken Coresight perf.data files
easier.

Signed-off-by: James Clark <james.clark@linaro.org>
---
Changes in v3:
- Remove old b4 prerequisite metadata that's no longer relevant.
- Link to v2: https://patch.msgid.link/20261002-james-cs-hw_id-output-failure-v2-0-7a194aa61173@linaro.org

Changes in v2:
- Drop driver changes and move to https://lore.kernel.org/linux-perf-users/20261001-james-cs-retry-hw_id-v1-0-4d45ab46cde7@linaro.org/T/#t
- Don't process any HW_IDs for unformatted queues, just create decoders
  unconditionally with trace ID 1.
- Use zfree() instead of etmq->decoder = NULL
- Remove dependency on "[PATCH v2 00/14] perf cs-etm: Per-thread mode
  fixes and snapshot wrap support", it wasn't functionally dependant,
  just touching some of the same lines.
- Link to v1: https://patch.msgid.link/20260821-james-cs-hw_id-output-failure-v1-0-9d532ddabcc3@linaro.org

---
James Clark (3):
      perf: cs-etm: Don't add global v0 HW_IDs to unformatted queues
      perf cs-etm: Free partially created queues
      perf: cs-etm: Respect --no-itrace option

Leo Yan (1):
      perf cs-etm: Synthesize missing HW_ID mappings for raw trace

 tools/perf/util/auxtrace.c                      |   2 +-
 tools/perf/util/auxtrace.h                      |   1 +
 tools/perf/util/cs-etm-decoder/cs-etm-decoder.c |  12 +-
 tools/perf/util/cs-etm-decoder/cs-etm-decoder.h |   2 +-
 tools/perf/util/cs-etm.c                        | 184 +++++++++++++++---------
 5 files changed, 123 insertions(+), 78 deletions(-)
---
base-commit: 705da5b15ab89ba97b11eedbe507c2fd83d31cb9
change-id: 20260706-james-cs-hw_id-output-failure-1d06d042ef96

Best regards,
--  
James Clark <james.clark@linaro.org>

[PATCH v3 1/4] perf: cs-etm: Don't add global v0 HW_IDs to unformatted queues

[James Clark]

In the case of mixed formatted and unformatted trace with v0 HW_IDs, or
no HW_IDs at all resulting in the cs_etm__map_trace_ids_metadata()
fallback, formatted queues will write their HW_IDs to unformatted
queues. This will result in an assert because of multiple decoders on an
unformatted queue.

Fix it by not inserting trace ID nodes on other unformatted queues. This
isn't an issue for v0.1 HW_IDs because they aren't global so will only
add to their own queue.

Fixes: 19c3e4db38c5 ("perf: cs-etm: Create decoders based on the trace ID mappings")
Signed-off-by: James Clark <james.clark@linaro.org>
---
 tools/perf/util/cs-etm.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c
index 2d1ab34f7b6b..fda6b6c02e26 100644
--- a/tools/perf/util/cs-etm.c
+++ b/tools/perf/util/cs-etm.c
@@ -309,27 +309,27 @@ static int cs_etm__map_trace_id_v0(struct cs_etm_auxtrace *etm, u8 trace_chan_id
 {
 	struct cs_etm_queue *etmq;
 
-	/*
-	 * If the queue is unformatted then only save one mapping in the
-	 * queue associated with that CPU so only one decoder is made.
-	 */
 	etmq = cs_etm__get_queue(etm, cpu_metadata[CS_ETM_CPU]);
 	if (!etmq)
 		return -EINVAL;
 
+	/*
+	 * If the queue is unformatted then only save one mapping in the
+	 * queue associated with that CPU so only one decoder is made.
+	 */
 	if (etmq->format == UNFORMATTED)
 		return cs_etm__insert_trace_id_node(etmq, trace_chan_id,
 						    cpu_metadata);
 
 	/*
 	 * Otherwise, version 0 trace IDs are global so save them into every
-	 * queue.
+	 * formatted queue.
 	 */
 	for (unsigned int i = 0; i < etm->queues.nr_queues; ++i) {
 		int ret;
 
 		etmq = etm->queues.queue_array[i].priv;
-		if (!etmq)
+		if (!etmq || etmq->format == UNFORMATTED)
 			continue;
 
 		ret = cs_etm__insert_trace_id_node(etmq, trace_chan_id,

-- 
2.43.0

Re: [PATCH v3 1/4] perf: cs-etm: Don't add global v0 HW_IDs to unformatted queues

[Leo Yan]

On Fri, Oct 02, 2026 at 12:15:11PM +0100, James Clark wrote:
> In the case of mixed formatted and unformatted trace with v0 HW_IDs, or
> no HW_IDs at all resulting in the cs_etm__map_trace_ids_metadata()
> fallback, formatted queues will write their HW_IDs to unformatted
> queues. This will result in an assert because of multiple decoders on an
> unformatted queue.
> 
> Fix it by not inserting trace ID nodes on other unformatted queues. This
> isn't an issue for v0.1 HW_IDs because they aren't global so will only
> add to their own queue.
> 
> Fixes: 19c3e4db38c5 ("perf: cs-etm: Create decoders based on the trace ID mappings")
> Signed-off-by: James Clark <james.clark@linaro.org>

Reviewed-by: Leo Yan <leo.yan@arm.com>

[PATCH v3 2/4] perf cs-etm: Free partially created queues

[James Clark]

cs_etm__create_decoders() can fail half way through and the
err_free_queues label won't free them. Call the full queue free function
here instead. If no decoders were created it's a nop.

Reviewed-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
---
 tools/perf/util/cs-etm-decoder/cs-etm-decoder.c | 12 ++++++------
 tools/perf/util/cs-etm-decoder/cs-etm-decoder.h |  2 +-
 tools/perf/util/cs-etm.c                        | 25 ++++++++++++++++---------
 3 files changed, 23 insertions(+), 16 deletions(-)

diff --git a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c
index 26940f1f1b0b..b56cf57a68c8 100644
--- a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c
+++ b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.c
@@ -741,7 +741,7 @@ cs_etm_decoder__new(int decoders, struct cs_etm_decoder_params *d_params,
 	return decoder;
 
 err_free_decoder:
-	cs_etm_decoder__free(decoder);
+	cs_etm_decoder__zfree(&decoder);
 	return NULL;
 }
 
@@ -793,14 +793,14 @@ int cs_etm_decoder__process_data_block(struct cs_etm_decoder *decoder,
 	return ret;
 }
 
-void cs_etm_decoder__free(struct cs_etm_decoder *decoder)
+void cs_etm_decoder__zfree(struct cs_etm_decoder **decoder)
 {
-	if (!decoder)
+	if (!(*decoder))
 		return;
 
-	ocsd_destroy_dcd_tree(decoder->dcd_tree);
-	decoder->dcd_tree = NULL;
-	free(decoder);
+	ocsd_destroy_dcd_tree((*decoder)->dcd_tree);
+	(*decoder)->dcd_tree = NULL;
+	zfree(decoder);
 }
 
 const char *cs_etm_decoder__get_name(struct cs_etm_decoder *decoder)
diff --git a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h
index 12c782fa6db2..2a8214f17958 100644
--- a/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h
+++ b/tools/perf/util/cs-etm-decoder/cs-etm-decoder.h
@@ -96,7 +96,7 @@ cs_etm_decoder__new(int num_cpu,
 		    struct cs_etm_decoder_params *d_params,
 		    struct cs_etm_trace_params t_params[]);
 
-void cs_etm_decoder__free(struct cs_etm_decoder *decoder);
+void cs_etm_decoder__zfree(struct cs_etm_decoder **decoder);
 
 int cs_etm_decoder__add_mem_access_cb(struct cs_etm_decoder *decoder,
 				      u64 start, u64 end,
diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c
index fda6b6c02e26..78ae8443ac96 100644
--- a/tools/perf/util/cs-etm.c
+++ b/tools/perf/util/cs-etm.c
@@ -1002,7 +1002,7 @@ static void cs_etm__free_queue(void *priv)
 	if (!etmq)
 		return;
 
-	cs_etm_decoder__free(etmq->decoder);
+	cs_etm_decoder__zfree(&etmq->decoder);
 	cs_etm__free_traceid_queues(etmq);
 
 	if (etmq->own_traceid_list) {
@@ -1017,13 +1017,10 @@ static void cs_etm__free_queue(void *priv)
 	free(etmq);
 }
 
-static void cs_etm__free_events(struct perf_session *session)
+static void cs_etm__free_queues(struct cs_etm_auxtrace *etm)
 {
 	unsigned int i;
-	struct cs_etm_auxtrace *aux = container_of(session->auxtrace,
-						   struct cs_etm_auxtrace,
-						   auxtrace);
-	struct auxtrace_queues *queues = &aux->queues;
+	struct auxtrace_queues *queues = &etm->queues;
 
 	for (i = 0; i < queues->nr_queues; i++) {
 		cs_etm__free_queue(queues->queue_array[i].priv);
@@ -1033,13 +1030,23 @@ static void cs_etm__free_events(struct perf_session *session)
 	auxtrace_queues__free(queues);
 }
 
+static void cs_etm__free_events(struct perf_session *session)
+{
+	struct cs_etm_auxtrace *aux = container_of(session->auxtrace,
+						   struct cs_etm_auxtrace,
+						   auxtrace);
+
+	cs_etm__free_queues(aux);
+}
+
 static void cs_etm__free(struct perf_session *session)
 {
 	int i;
 	struct cs_etm_auxtrace *aux = container_of(session->auxtrace,
 						   struct cs_etm_auxtrace,
 						   auxtrace);
-	cs_etm__free_events(session);
+
+	cs_etm__free_queues(aux);
 	session->auxtrace = NULL;
 
 	for (i = 0; i < aux->num_cpu; i++)
@@ -3476,7 +3483,7 @@ static int cs_etm__create_queue_decoders(struct cs_etm_queue *etmq)
 	return 0;
 
 out_free_decoder:
-	cs_etm_decoder__free(etmq->decoder);
+	cs_etm_decoder__zfree(&etmq->decoder);
 out_free:
 	zfree(&t_params);
 	return -EINVAL;
@@ -3758,7 +3765,7 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event,
 	return 0;
 
 err_free_queues:
-	auxtrace_queues__free(&etm->queues);
+	cs_etm__free_queues(etm);
 	session->auxtrace = NULL;
 err_free_etm:
 	zfree(&etm);

-- 
2.43.0

[PATCH v3 3/4] perf cs-etm: Synthesize missing HW_ID mappings for raw trace

[James Clark]

From: Leo Yan <leo.yan@arm.com>

The CoreSight driver emits PERF_RECORD_AUX_OUTPUT_HW_ID via
perf_report_aux_output_id(). If the perf ring buffer is full,
perf_report_aux_output_id() can fail to emit the record and return
silently.

When at least one HW_ID record is present in the file, perf skips the
legacy metadata-only trace ID mapping path. Therefore any raw AUX queue
whose HW_ID record was dropped can be left with an empty traceid_list,
and cs_etm__create_queue_decoders() will not create a decoder for that
queue.

Fix it by ignoring all HW_IDs for unformatted queues and always create
one decoder per queue unconditionally. Trace ID is unused so we can hard
code it to 1 (0 is a special value used in
'pending_timestamp_chan_id' to indicate not pending). We also need put
the same trace ID in the CPU metadata because the relationship between
CPU metadata and traceid_list is bidirectional. Moving
cs_etm__metadata_set_trace_id() into the mapping function solves this
and simplifies existing code. cs_etm__insert_trace_id_node() now does
two things, so rename it to a more generic cs_etm__map_trace_id().

Signed-off-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
---
 tools/perf/util/cs-etm.c | 85 +++++++++++++++++++++++++++++++++---------------
 1 file changed, 58 insertions(+), 27 deletions(-)

diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c
index 78ae8443ac96..c0152549e3e2 100644
--- a/tools/perf/util/cs-etm.c
+++ b/tools/perf/util/cs-etm.c
@@ -243,8 +243,12 @@ enum cs_etm_pid_fmt cs_etm__get_pid_fmt(struct cs_etm_queue *etmq)
 	return etmq->etm->pid_fmt;
 }
 
-static int cs_etm__insert_trace_id_node(struct cs_etm_queue *etmq,
-					u8 trace_chan_id, u64 *cpu_metadata)
+/*
+ * Add a traceid_list entry that maps a trace ID to a CPU metadata for this etmq
+ * and update the trace ID in the cpu_metadata.
+ */
+static int cs_etm__map_trace_id(struct cs_etm_queue *etmq, u8 trace_chan_id,
+				u64 *cpu_metadata)
 {
 	/* Get an RB node for this CPU */
 	struct int_node *inode = intlist__findnew(etmq->traceid_list, trace_chan_id);
@@ -289,8 +293,7 @@ static int cs_etm__insert_trace_id_node(struct cs_etm_queue *etmq,
 
 	/* Not one we've seen before, associate the traceID with the metadata pointer */
 	inode->priv = cpu_metadata;
-
-	return 0;
+	return cs_etm__metadata_set_trace_id(trace_chan_id, cpu_metadata);
 }
 
 static struct cs_etm_queue *cs_etm__get_queue(struct cs_etm_auxtrace *etm, int cpu)
@@ -314,12 +317,11 @@ static int cs_etm__map_trace_id_v0(struct cs_etm_auxtrace *etm, u8 trace_chan_id
 		return -EINVAL;
 
 	/*
-	 * If the queue is unformatted then only save one mapping in the
-	 * queue associated with that CPU so only one decoder is made.
+	 * Ignore HW_IDs for unformatted, a decoder will be made
+	 * unconditionally later.
 	 */
 	if (etmq->format == UNFORMATTED)
-		return cs_etm__insert_trace_id_node(etmq, trace_chan_id,
-						    cpu_metadata);
+		return 0;
 
 	/*
 	 * Otherwise, version 0 trace IDs are global so save them into every
@@ -332,8 +334,7 @@ static int cs_etm__map_trace_id_v0(struct cs_etm_auxtrace *etm, u8 trace_chan_id
 		if (!etmq || etmq->format == UNFORMATTED)
 			continue;
 
-		ret = cs_etm__insert_trace_id_node(etmq, trace_chan_id,
-						   cpu_metadata);
+		ret = cs_etm__map_trace_id(etmq, trace_chan_id, cpu_metadata);
 		if (ret)
 			return ret;
 	}
@@ -344,7 +345,6 @@ static int cs_etm__map_trace_id_v0(struct cs_etm_auxtrace *etm, u8 trace_chan_id
 static int cs_etm__process_trace_id_v0(struct cs_etm_auxtrace *etm, int cpu,
 				       u64 hw_id)
 {
-	int err;
 	u64 *cpu_data;
 	u8 trace_chan_id = FIELD_GET(CS_AUX_HW_ID_TRACE_ID_MASK, hw_id);
 
@@ -352,22 +352,13 @@ static int cs_etm__process_trace_id_v0(struct cs_etm_auxtrace *etm, int cpu,
 	if (cpu_data == NULL)
 		return -EINVAL;
 
-	err = cs_etm__map_trace_id_v0(etm, trace_chan_id, cpu_data);
-	if (err)
-		return err;
-
-	/*
-	 * if we are picking up the association from the packet, need to plug
-	 * the correct trace ID into the metadata for setting up decoders later.
-	 */
-	return cs_etm__metadata_set_trace_id(trace_chan_id, cpu_data);
+	return cs_etm__map_trace_id_v0(etm, trace_chan_id, cpu_data);
 }
 
 static int cs_etm__process_trace_id_v0_1(struct cs_etm_auxtrace *etm, int cpu,
 					 u64 hw_id)
 {
 	struct cs_etm_queue *etmq = cs_etm__get_queue(etm, cpu);
-	int ret;
 	u64 *cpu_data;
 	u32 sink_id = FIELD_GET(CS_AUX_HW_ID_SINK_ID_MASK, hw_id);
 	u8 trace_id = FIELD_GET(CS_AUX_HW_ID_TRACE_ID_MASK, hw_id);
@@ -375,6 +366,13 @@ static int cs_etm__process_trace_id_v0_1(struct cs_etm_auxtrace *etm, int cpu,
 	if (!etmq)
 		return -EINVAL;
 
+	/*
+	 * Ignore HW_IDs for unformatted, a decoder will be made
+	 * unconditionally later.
+	 */
+	if (etmq->format == UNFORMATTED)
+		return 0;
+
 	/*
 	 * Check sink id hasn't changed in per-cpu mode. In per-thread mode,
 	 * let it pass for now until an actual overlapping trace ID is hit. In
@@ -419,13 +417,41 @@ static int cs_etm__process_trace_id_v0_1(struct cs_etm_auxtrace *etm, int cpu,
 	if (!cpu_data)
 		return -EINVAL;
 
-	ret = cs_etm__insert_trace_id_node(etmq, trace_id, cpu_data);
-	if (ret)
-		return ret;
+	return cs_etm__map_trace_id(etmq, trace_id, cpu_data);
+}
 
-	ret = cs_etm__metadata_set_trace_id(trace_id, cpu_data);
-	if (ret)
-		return ret;
+static int cs_etm__map_unformatted_trace_ids(struct cs_etm_auxtrace *etm)
+{
+	struct auxtrace_queues *queues = &etm->queues;
+
+	for (unsigned int i = 0; i < queues->nr_queues; i++) {
+		struct auxtrace_queue *queue = &queues->queue_array[i];
+		struct cs_etm_queue *etmq = queue->priv;
+		u64 *cpu_data;
+		int ret;
+
+		if (list_empty(&queue->head) || !etmq || etmq->format != UNFORMATTED)
+			continue;
+
+		if (!intlist__empty(etmq->traceid_list)) {
+			pr_err("CS_ETM: Unformatted queue should't have trace ID\n");
+			return -EINVAL;
+		}
+
+		if (etm->per_thread_decoding)
+			cpu_data = etm->metadata[0];
+		else
+			cpu_data = get_cpu_data(etm, queue->cpu);
+
+		if (!cpu_data) {
+			pr_err("CS_ETM: Unformatted queue missing CPU metadata\n");
+			return -EINVAL;
+		}
+
+		ret = cs_etm__map_trace_id(etmq, 1, cpu_data);
+		if (ret)
+			return ret;
+	}
 
 	return 0;
 }
@@ -3757,6 +3783,11 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event,
 			goto err_free_queues;
 	}
 
+	/* Create IDs for TRBE sinks so that a decoder is made for each one. */
+	err = cs_etm__map_unformatted_trace_ids(etm);
+	if (err)
+		goto err_free_queues;
+
 	err = cs_etm__create_decoders(etm);
 	if (err)
 		goto err_free_queues;

-- 
2.43.0

Re: [PATCH v3 3/4] perf cs-etm: Synthesize missing HW_ID mappings for raw trace

[Leo Yan]

On Fri, Oct 02, 2026 at 12:15:13PM +0100, James Clark wrote:

[...]

>  static struct cs_etm_queue *cs_etm__get_queue(struct cs_etm_auxtrace *etm, int cpu)
> @@ -314,12 +317,11 @@ static int cs_etm__map_trace_id_v0(struct cs_etm_auxtrace *etm, u8 trace_chan_id
>  		return -EINVAL;
>  
>  	/*
> -	 * If the queue is unformatted then only save one mapping in the
> -	 * queue associated with that CPU so only one decoder is made.
> +	 * Ignore HW_IDs for unformatted, a decoder will be made
> +	 * unconditionally later.
>  	 */
>  	if (etmq->format == UNFORMATTED)
> -		return cs_etm__insert_trace_id_node(etmq, trace_chan_id,
> -						    cpu_metadata);
> +		return 0;

Nitpick: It seems to me that we can move the
etmq->format == UNFORMATTED check into cs_etm__process_aux_output_hw_id().

This avoids duplicating the check in both cs_etm__map_trace_id_v0()
and cs_etm__process_trace_id_v0_1(), and makes it more apparent that
the unformatted case does not use HW_ID at all.

It's up to you whether to keep it as is or update it. Either way:

Reviewed-by: Leo Yan leo.yan@arm.com

P.S. Although I contributed the initial idea, the change is now quite
different from the original patch I shared. Please feel free to drop my
name if you prefer.

[PATCH v3 4/4] perf: cs-etm: Respect --no-itrace option

[James Clark]

If there is a bug in decoding then Perf will exit early and you can't
investigate by using the raw dump mode. Make the --no-itrace option stop
Coresight decoding after printing the aux header so decode errors don't
stop the rest of the samples being printed.

pmu_type and evsel_is_auxtrace need to be set so that the
unleader_auxtrace() behavior is the same as without --no-itrace.

Reviewed-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: James Clark <james.clark@linaro.org>
---
 tools/perf/util/auxtrace.c |  2 +-
 tools/perf/util/auxtrace.h |  1 +
 tools/perf/util/cs-etm.c   | 66 +++++++++++++++++++++++++---------------------
 3 files changed, 38 insertions(+), 31 deletions(-)

diff --git a/tools/perf/util/auxtrace.c b/tools/perf/util/auxtrace.c
index ed2113190785..2557cbec37b3 100644
--- a/tools/perf/util/auxtrace.c
+++ b/tools/perf/util/auxtrace.c
@@ -123,7 +123,7 @@ static int evlist__regroup(struct evlist *evlist, struct evsel *leader, struct e
 	return 0;
 }
 
-static bool auxtrace__dont_decode(struct perf_session *session)
+bool auxtrace__dont_decode(struct perf_session *session)
 {
 	return !session->itrace_synth_opts ||
 	       session->itrace_synth_opts->dont_decode;
diff --git a/tools/perf/util/auxtrace.h b/tools/perf/util/auxtrace.h
index 5012bead1b92..c29a8444f08e 100644
--- a/tools/perf/util/auxtrace.h
+++ b/tools/perf/util/auxtrace.h
@@ -650,6 +650,7 @@ void auxtrace__free(struct perf_session *session);
 bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
 				 struct evsel *evsel);
 u64 auxtrace_synth_id_range_start(struct evsel *evsel);
+bool auxtrace__dont_decode(struct perf_session *session);
 
 #define ITRACE_HELP \
 "				i[period]:    		synthesize instructions events\n" \
diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c
index c0152549e3e2..bdcf94813129 100644
--- a/tools/perf/util/cs-etm.c
+++ b/tools/perf/util/cs-etm.c
@@ -965,7 +965,7 @@ static int cs_etm__flush_events(struct perf_session *session,
 	struct cs_etm_auxtrace *etm = container_of(session->auxtrace,
 						   struct cs_etm_auxtrace,
 						   auxtrace);
-	if (dump_trace)
+	if (dump_trace || auxtrace__dont_decode(session))
 		return 0;
 
 	if (!tool->ordered_events)
@@ -3002,7 +3002,7 @@ static int cs_etm__process_event(struct perf_session *session,
 						   struct cs_etm_auxtrace,
 						   auxtrace);
 
-	if (dump_trace)
+	if (dump_trace || auxtrace__dont_decode(session))
 		return 0;
 
 	if (!tool->ordered_events) {
@@ -3552,27 +3552,49 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event,
 	int err = 0;
 	int aux_hw_id_found;
 	int i;
-	u64 *ptr = NULL;
+	u64 *ptr = (u64 *) auxtrace_info->priv;
 	u64 **metadata = NULL;
 
-	/* First the global part */
-	ptr = (u64 *) auxtrace_info->priv;
-	num_cpu = ptr[CS_PMU_TYPE_CPUS] & 0xffffffff;
+	etm = zalloc(sizeof(*etm));
+	if (!etm)
+		return -ENOMEM;
+
+	session->auxtrace = &etm->auxtrace;
+	etm->auxtrace.free = cs_etm__free;
+	etm->auxtrace.evsel_is_auxtrace = cs_etm__evsel_is_auxtrace;
+	etm->auxtrace.process_event = cs_etm__process_event;
+	etm->auxtrace.process_auxtrace_event = cs_etm__process_auxtrace_event;
+	etm->auxtrace.flush_events = cs_etm__flush_events;
+	etm->auxtrace.free_events = cs_etm__free_events;
+	etm->pmu_type = (unsigned int) ((ptr[CS_PMU_TYPE_CPUS] >> 32) & 0xffffffff);
+
+	/*
+	 * Don't go further than the minimum required to identify this event as
+	 * auxtrace with cs_etm__evsel_is_auxtrace() so unleader_auxtrace()
+	 * works.
+	 */
+	if (auxtrace__dont_decode(session))
+		return 0;
 
 	/*
 	 * Bound num_cpu by the event size: the global header consumes
 	 * CS_ETM_HEADER_SIZE bytes, and each CPU needs at least one u64
 	 * metadata entry after that.
 	 */
+	num_cpu = ptr[CS_PMU_TYPE_CPUS] & 0xffffffff;
 	priv_size = total_size - event_header_size - INFO_HEADER_SIZE -
 		    CS_ETM_HEADER_SIZE;
 	if (num_cpu <= 0 || priv_size <= 0 ||
-	    num_cpu > priv_size / (int)sizeof(u64))
-		return -EINVAL;
+	    num_cpu > priv_size / (int)sizeof(u64)) {
+		err = -EINVAL;
+		goto err_free_etm;
+	}
 
 	metadata = zalloc(sizeof(*metadata) * num_cpu);
-	if (!metadata)
-		return -ENOMEM;
+	if (!metadata) {
+		err = -ENOMEM;
+		goto err_free_etm;
+	}
 
 	/* Start parsing after the common part of the header */
 	i = CS_HEADER_VERSION_MAX;
@@ -3631,13 +3653,6 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event,
 		goto err_free_metadata;
 	}
 
-	etm = zalloc(sizeof(*etm));
-
-	if (!etm) {
-		err = -ENOMEM;
-		goto err_free_metadata;
-	}
-
 	/*
 	 * As all the ETMs run at the same exception level, the system should
 	 * have the same PID format crossing CPUs.  So cache the PID format
@@ -3647,7 +3662,7 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event,
 
 	err = auxtrace_queues__init_nr(&etm->queues, max_cpu + 1);
 	if (err)
-		goto err_free_etm;
+		goto err_free_metadata;
 
 	for (unsigned int j = 0; j < etm->queues.nr_queues; ++j) {
 		err = cs_etm__setup_queue(etm, &etm->queues.queue_array[j], j);
@@ -3685,7 +3700,6 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event,
 	etm->session = session;
 
 	etm->num_cpu = num_cpu;
-	etm->pmu_type = (unsigned int) ((ptr[CS_PMU_TYPE_CPUS] >> 32) & 0xffffffff);
 	etm->snapshot_mode = (ptr[CS_ETM_SNAPSHOT] != 0);
 	etm->metadata = metadata;
 	etm->auxtrace_type = auxtrace_info->type;
@@ -3712,14 +3726,6 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event,
 			    "you can specify the itrace option 'T' for timestamp decoding\n"
 			    "if the Coresight timestamp on the platform is same with the kernel time.\n\n");
 
-	etm->auxtrace.process_event = cs_etm__process_event;
-	etm->auxtrace.process_auxtrace_event = cs_etm__process_auxtrace_event;
-	etm->auxtrace.flush_events = cs_etm__flush_events;
-	etm->auxtrace.free_events = cs_etm__free_events;
-	etm->auxtrace.free = cs_etm__free;
-	etm->auxtrace.evsel_is_auxtrace = cs_etm__evsel_is_auxtrace;
-	session->auxtrace = &etm->auxtrace;
-
 	cs_etm__setup_timeless_decoding(etm);
 
 	etm->tc.time_shift = tc->time_shift;
@@ -3797,13 +3803,13 @@ int cs_etm__process_auxtrace_info_full(union perf_event *event,
 
 err_free_queues:
 	cs_etm__free_queues(etm);
-	session->auxtrace = NULL;
-err_free_etm:
-	zfree(&etm);
 err_free_metadata:
 	/* No need to check @metadata[j], free(NULL) is supported */
 	for (int j = 0; j < num_cpu; j++)
 		zfree(&metadata[j]);
 	zfree(&metadata);
+err_free_etm:
+	session->auxtrace = NULL;
+	zfree(&etm);
 	return err;
 }

-- 
2.43.0

Re: [PATCH v3 4/4] perf: cs-etm: Respect --no-itrace option

[Leo Yan]

On Fri, Oct 02, 2026 at 12:15:14PM +0100, James Clark wrote:
> If there is a bug in decoding then Perf will exit early and you can't
> investigate by using the raw dump mode. Make the --no-itrace option stop
> Coresight decoding after printing the aux header so decode errors don't
> stop the rest of the samples being printed.
> 
> pmu_type and evsel_is_auxtrace need to be set so that the
> unleader_auxtrace() behavior is the same as without --no-itrace.
> 
> Reviewed-by: Adrian Hunter <adrian.hunter@intel.com>
> Signed-off-by: James Clark <james.clark@linaro.org>

Reviewed-by: Leo Yan <leo.yan@arm.com>