[PATCH v4 0/5] perf arm-spe: Introduce metadata version 2

[PATCH v4 0/5] perf arm-spe: Introduce metadata version 2

[Leo Yan]

This patch series enhances Arm SPE metadata in the Perf file to a
version 2 format and maintains backward compatibility for metadata v1.

The version 2 metadata contains a global structure with fields for
metadata header version number, header size, CPU numbers. It also
expands to store per-CPU metadata, which includes the CPU logical
number in the kernel, MIDR, and capacity values associated with Arm SPE.

This patch set has been tested the perf to decode the Arm SPE metadata
v1 and v2.

Changes from v3:
- Fixed a compilation error (James)
- Added James' review tags.

Changes from v2:
- Added comment for arm_spe_find_cpus (Namhyung)
- Corrected resource releasing in arm_spe_save_cpu_header() and
  arm_spe_info_fill() (Namhyung)
- Changed to use calloc() for allocating metadata pointer array
  (Namhyung)

Changes from v1:
- Dropped LDS bit exposing from Arm SPE driver (Will Deacon).
- To simplify the change, this series did not include multiple AUX event
  support.

Leo Yan (5):
  perf arm-spe: Define metadata header version 2
  perf arm-spe: Calculate meta data size
  perf arm-spe: Save per CPU information in metadata
  perf arm-spe: Support metadata version 2
  perf arm-spe: Dump metadata with version 2

 tools/perf/arch/arm64/util/arm-spe.c | 122 ++++++++++++++++++++-
 tools/perf/util/arm-spe.c            | 155 +++++++++++++++++++++++++--
 tools/perf/util/arm-spe.h            |  38 ++++++-
 3 files changed, 299 insertions(+), 16 deletions(-)

-- 
2.34.1

[PATCH v4 1/5] perf arm-spe: Define metadata header version 2

[Leo Yan]

The first version's metadata header structure doesn't include a field to
indicate a header version, which is not friendly for extension.

Define the metadata version 2 format with a new header structure and
extend per CPU's metadata. In the meantime, the old metadata header will
still be supported for backward compatibility.

Signed-off-by: Leo Yan <leo.yan@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
---
 tools/perf/arch/arm64/util/arm-spe.c |  4 +--
 tools/perf/util/arm-spe.c            |  2 +-
 tools/perf/util/arm-spe.h            | 38 +++++++++++++++++++++++++++-
 3 files changed, 40 insertions(+), 4 deletions(-)

diff --git a/tools/perf/arch/arm64/util/arm-spe.c b/tools/perf/arch/arm64/util/arm-spe.c
index 2be99fdf997d..c2d5c8ca4900 100644
--- a/tools/perf/arch/arm64/util/arm-spe.c
+++ b/tools/perf/arch/arm64/util/arm-spe.c
@@ -41,7 +41,7 @@ static size_t
 arm_spe_info_priv_size(struct auxtrace_record *itr __maybe_unused,
 		       struct evlist *evlist __maybe_unused)
 {
-	return ARM_SPE_AUXTRACE_PRIV_SIZE;
+	return ARM_SPE_AUXTRACE_V1_PRIV_SIZE;
 }
 
 static int arm_spe_info_fill(struct auxtrace_record *itr,
@@ -53,7 +53,7 @@ static int arm_spe_info_fill(struct auxtrace_record *itr,
 			container_of(itr, struct arm_spe_recording, itr);
 	struct perf_pmu *arm_spe_pmu = sper->arm_spe_pmu;
 
-	if (priv_size != ARM_SPE_AUXTRACE_PRIV_SIZE)
+	if (priv_size != ARM_SPE_AUXTRACE_V1_PRIV_SIZE)
 		return -EINVAL;
 
 	if (!session->evlist->core.nr_mmaps)
diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
index 138ffc71b32d..70989b1bae47 100644
--- a/tools/perf/util/arm-spe.c
+++ b/tools/perf/util/arm-spe.c
@@ -1262,7 +1262,7 @@ int arm_spe_process_auxtrace_info(union perf_event *event,
 				  struct perf_session *session)
 {
 	struct perf_record_auxtrace_info *auxtrace_info = &event->auxtrace_info;
-	size_t min_sz = sizeof(u64) * ARM_SPE_AUXTRACE_PRIV_MAX;
+	size_t min_sz = ARM_SPE_AUXTRACE_V1_PRIV_SIZE;
 	struct perf_record_time_conv *tc = &session->time_conv;
 	const char *cpuid = perf_env__cpuid(session->evlist->env);
 	u64 midr = strtol(cpuid, NULL, 16);
diff --git a/tools/perf/util/arm-spe.h b/tools/perf/util/arm-spe.h
index 4f4900c18f3e..390679a4af2f 100644
--- a/tools/perf/util/arm-spe.h
+++ b/tools/perf/util/arm-spe.h
@@ -12,10 +12,46 @@
 enum {
 	ARM_SPE_PMU_TYPE,
 	ARM_SPE_PER_CPU_MMAPS,
+	ARM_SPE_AUXTRACE_V1_PRIV_MAX,
+};
+
+#define ARM_SPE_AUXTRACE_V1_PRIV_SIZE	\
+	(ARM_SPE_AUXTRACE_V1_PRIV_MAX * sizeof(u64))
+
+enum {
+	/*
+	 * The old metadata format (defined above) does not include a
+	 * field for version number. Version 1 is reserved and starts
+	 * from version 2.
+	 */
+	ARM_SPE_HEADER_VERSION,
+	/* Number of sizeof(u64) */
+	ARM_SPE_HEADER_SIZE,
+	/* PMU type shared by CPUs */
+	ARM_SPE_PMU_TYPE_V2,
+	/* Number of CPUs */
+	ARM_SPE_CPUS_NUM,
 	ARM_SPE_AUXTRACE_PRIV_MAX,
 };
 
-#define ARM_SPE_AUXTRACE_PRIV_SIZE (ARM_SPE_AUXTRACE_PRIV_MAX * sizeof(u64))
+enum {
+	/* Magic number */
+	ARM_SPE_MAGIC,
+	/* CPU logical number in system */
+	ARM_SPE_CPU,
+	/* Number of parameters */
+	ARM_SPE_CPU_NR_PARAMS,
+	/* CPU MIDR */
+	ARM_SPE_CPU_MIDR,
+	/* Associated PMU type */
+	ARM_SPE_CPU_PMU_TYPE,
+	/* Minimal interval */
+	ARM_SPE_CAP_MIN_IVAL,
+	ARM_SPE_CPU_PRIV_MAX,
+};
+
+#define ARM_SPE_HEADER_CURRENT_VERSION	2
+
 
 union perf_event;
 struct perf_session;
-- 
2.34.1

[PATCH v4 2/5] perf arm-spe: Calculate meta data size

[Leo Yan]

The metadata is designed to contain a header and per CPU information.

The arm_spe_find_cpus() function is introduced to identify how many CPUs
support ARM SPE. Based on the CPU number, calculates the metadata size.

Signed-off-by: Leo Yan <leo.yan@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
---
 tools/perf/arch/arm64/util/arm-spe.c | 39 +++++++++++++++++++++++++---
 1 file changed, 36 insertions(+), 3 deletions(-)

diff --git a/tools/perf/arch/arm64/util/arm-spe.c b/tools/perf/arch/arm64/util/arm-spe.c
index c2d5c8ca4900..81d5c07380a4 100644
--- a/tools/perf/arch/arm64/util/arm-spe.c
+++ b/tools/perf/arch/arm64/util/arm-spe.c
@@ -37,11 +37,44 @@ struct arm_spe_recording {
 	bool			*wrapped;
 };
 
+/*
+ * arm_spe_find_cpus() returns a new cpu map, and the caller should invoke
+ * perf_cpu_map__put() to release the map after use.
+ */
+static struct perf_cpu_map *arm_spe_find_cpus(struct evlist *evlist)
+{
+	struct perf_cpu_map *event_cpus = evlist->core.user_requested_cpus;
+	struct perf_cpu_map *online_cpus = perf_cpu_map__new_online_cpus();
+	struct perf_cpu_map *intersect_cpus;
+
+	/* cpu map is not "any" CPU , we have specific CPUs to work with */
+	if (!perf_cpu_map__has_any_cpu(event_cpus)) {
+		intersect_cpus = perf_cpu_map__intersect(event_cpus, online_cpus);
+		perf_cpu_map__put(online_cpus);
+	/* Event can be "any" CPU so count all CPUs. */
+	} else {
+		intersect_cpus = online_cpus;
+	}
+
+	return intersect_cpus;
+}
+
 static size_t
 arm_spe_info_priv_size(struct auxtrace_record *itr __maybe_unused,
-		       struct evlist *evlist __maybe_unused)
+		       struct evlist *evlist)
 {
-	return ARM_SPE_AUXTRACE_V1_PRIV_SIZE;
+	struct perf_cpu_map *cpu_map = arm_spe_find_cpus(evlist);
+	size_t size;
+
+	if (!cpu_map)
+		return 0;
+
+	size = ARM_SPE_AUXTRACE_PRIV_MAX +
+	       ARM_SPE_CPU_PRIV_MAX * perf_cpu_map__nr(cpu_map);
+	size *= sizeof(u64);
+
+	perf_cpu_map__put(cpu_map);
+	return size;
 }
 
 static int arm_spe_info_fill(struct auxtrace_record *itr,
@@ -53,7 +86,7 @@ static int arm_spe_info_fill(struct auxtrace_record *itr,
 			container_of(itr, struct arm_spe_recording, itr);
 	struct perf_pmu *arm_spe_pmu = sper->arm_spe_pmu;
 
-	if (priv_size != ARM_SPE_AUXTRACE_V1_PRIV_SIZE)
+	if (priv_size != arm_spe_info_priv_size(itr, session->evlist))
 		return -EINVAL;
 
 	if (!session->evlist->core.nr_mmaps)
-- 
2.34.1

[PATCH v4 3/5] perf arm-spe: Save per CPU information in metadata

[Leo Yan]

Save the Arm SPE information on a per-CPU basis. This approach is easier
in the decoding phase for retrieving metadata based on the CPU number of
every Arm SPE record.

Signed-off-by: Leo Yan <leo.yan@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
---
 tools/perf/arch/arm64/util/arm-spe.c | 83 +++++++++++++++++++++++++++-
 1 file changed, 81 insertions(+), 2 deletions(-)

diff --git a/tools/perf/arch/arm64/util/arm-spe.c b/tools/perf/arch/arm64/util/arm-spe.c
index 81d5c07380a4..28b35463d274 100644
--- a/tools/perf/arch/arm64/util/arm-spe.c
+++ b/tools/perf/arch/arm64/util/arm-spe.c
@@ -26,6 +26,8 @@
 #include "../../../util/arm-spe.h"
 #include <tools/libc_compat.h> // reallocarray
 
+#define ARM_SPE_CPU_MAGIC		0x1010101010101010ULL
+
 #define KiB(x) ((x) * 1024)
 #define MiB(x) ((x) * 1024 * 1024)
 
@@ -77,14 +79,70 @@ arm_spe_info_priv_size(struct auxtrace_record *itr __maybe_unused,
 	return size;
 }
 
+static int arm_spe_save_cpu_header(struct auxtrace_record *itr,
+				   struct perf_cpu cpu, __u64 data[])
+{
+	struct arm_spe_recording *sper =
+			container_of(itr, struct arm_spe_recording, itr);
+	struct perf_pmu *pmu = NULL;
+	struct perf_pmu tmp_pmu;
+	char cpu_id_str[16];
+	char *cpuid = NULL;
+	u64 val;
+
+	snprintf(cpu_id_str, sizeof(cpu_id_str), "%d", cpu.cpu);
+	tmp_pmu.cpus = perf_cpu_map__new(cpu_id_str);
+	if (!tmp_pmu.cpus)
+		return -ENOMEM;
+
+	/* Read CPU MIDR */
+	cpuid = perf_pmu__getcpuid(&tmp_pmu);
+
+	/* The CPU map will not be used anymore, release it */
+	perf_cpu_map__put(tmp_pmu.cpus);
+
+	if (!cpuid)
+		return -ENOMEM;
+	val = strtol(cpuid, NULL, 16);
+
+	data[ARM_SPE_MAGIC] = ARM_SPE_CPU_MAGIC;
+	data[ARM_SPE_CPU] = cpu.cpu;
+	data[ARM_SPE_CPU_NR_PARAMS] = ARM_SPE_CPU_PRIV_MAX - ARM_SPE_CPU_MIDR;
+	data[ARM_SPE_CPU_MIDR] = val;
+
+	/* Find the associate Arm SPE PMU for the CPU */
+	if (perf_cpu_map__has(sper->arm_spe_pmu->cpus, cpu))
+		pmu = sper->arm_spe_pmu;
+
+	if (!pmu) {
+		/* No Arm SPE PMU is found */
+		data[ARM_SPE_CPU_PMU_TYPE] = ULLONG_MAX;
+		data[ARM_SPE_CAP_MIN_IVAL] = 0;
+	} else {
+		data[ARM_SPE_CPU_PMU_TYPE] = pmu->type;
+
+		if (perf_pmu__scan_file(pmu, "caps/min_interval", "%lu", &val) != 1)
+			val = 0;
+		data[ARM_SPE_CAP_MIN_IVAL] = val;
+	}
+
+	free(cpuid);
+	return ARM_SPE_CPU_PRIV_MAX;
+}
+
 static int arm_spe_info_fill(struct auxtrace_record *itr,
 			     struct perf_session *session,
 			     struct perf_record_auxtrace_info *auxtrace_info,
 			     size_t priv_size)
 {
+	int i, ret;
+	size_t offset;
 	struct arm_spe_recording *sper =
 			container_of(itr, struct arm_spe_recording, itr);
 	struct perf_pmu *arm_spe_pmu = sper->arm_spe_pmu;
+	struct perf_cpu_map *cpu_map;
+	struct perf_cpu cpu;
+	__u64 *data;
 
 	if (priv_size != arm_spe_info_priv_size(itr, session->evlist))
 		return -EINVAL;
@@ -92,10 +150,31 @@ static int arm_spe_info_fill(struct auxtrace_record *itr,
 	if (!session->evlist->core.nr_mmaps)
 		return -EINVAL;
 
+	cpu_map = arm_spe_find_cpus(session->evlist);
+	if (!cpu_map)
+		return -EINVAL;
+
 	auxtrace_info->type = PERF_AUXTRACE_ARM_SPE;
-	auxtrace_info->priv[ARM_SPE_PMU_TYPE] = arm_spe_pmu->type;
+	auxtrace_info->priv[ARM_SPE_HEADER_VERSION] = ARM_SPE_HEADER_CURRENT_VERSION;
+	auxtrace_info->priv[ARM_SPE_HEADER_SIZE] =
+		ARM_SPE_AUXTRACE_PRIV_MAX - ARM_SPE_HEADER_VERSION;
+	auxtrace_info->priv[ARM_SPE_PMU_TYPE_V2] = arm_spe_pmu->type;
+	auxtrace_info->priv[ARM_SPE_CPUS_NUM] = perf_cpu_map__nr(cpu_map);
+
+	offset = ARM_SPE_AUXTRACE_PRIV_MAX;
+	perf_cpu_map__for_each_cpu(cpu, i, cpu_map) {
+		assert(offset < priv_size);
+		data = &auxtrace_info->priv[offset];
+		ret = arm_spe_save_cpu_header(itr, cpu, data);
+		if (ret < 0)
+			goto out;
+		offset += ret;
+	}
 
-	return 0;
+	ret = 0;
+out:
+	perf_cpu_map__put(cpu_map);
+	return ret;
 }
 
 static void
-- 
2.34.1

[PATCH v4 4/5] perf arm-spe: Support metadata version 2

[Leo Yan]

This commit is to support metadata version 2 and at the meantime it is
backward compatible for version 1's format.

The metadata version 1 doesn't include the ARM_SPE_HEADER_VERSION field.
As version 1 is fixed with two u64 fields, by checking the metadata
size, it distinguishes the metadata is version 1 or version 2 (and any
new versions if later will have). For version 2, it reads out CPU number
and retrieves the metadata info for every CPU.

Signed-off-by: Leo Yan <leo.yan@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
---
 tools/perf/util/arm-spe.c | 99 +++++++++++++++++++++++++++++++++++++--
 1 file changed, 95 insertions(+), 4 deletions(-)

diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
index 70989b1bae47..8b2af1c7dc9b 100644
--- a/tools/perf/util/arm-spe.c
+++ b/tools/perf/util/arm-spe.c
@@ -78,6 +78,10 @@ struct arm_spe {
 
 	unsigned long			num_events;
 	u8				use_ctx_pkt_for_pid;
+
+	u64				**metadata;
+	u64				metadata_ver;
+	u64				metadata_nr_cpu;
 };
 
 struct arm_spe_queue {
@@ -1016,6 +1020,73 @@ static int arm_spe_flush(struct perf_session *session __maybe_unused,
 	return 0;
 }
 
+static u64 *arm_spe__alloc_per_cpu_metadata(u64 *buf, int per_cpu_size)
+{
+	u64 *metadata;
+
+	metadata = zalloc(per_cpu_size);
+	if (!metadata)
+		return NULL;
+
+	memcpy(metadata, buf, per_cpu_size);
+	return metadata;
+}
+
+static void arm_spe__free_metadata(u64 **metadata, int nr_cpu)
+{
+	int i;
+
+	for (i = 0; i < nr_cpu; i++)
+		zfree(&metadata[i]);
+	free(metadata);
+}
+
+static u64 **arm_spe__alloc_metadata(struct perf_record_auxtrace_info *info,
+				     u64 *ver, int *nr_cpu)
+{
+	u64 *ptr = (u64 *)info->priv;
+	u64 metadata_size;
+	u64 **metadata = NULL;
+	int hdr_sz, per_cpu_sz, i;
+
+	metadata_size = info->header.size -
+		sizeof(struct perf_record_auxtrace_info);
+
+	/* Metadata version 1 */
+	if (metadata_size == ARM_SPE_AUXTRACE_V1_PRIV_SIZE) {
+		*ver = 1;
+		*nr_cpu = 0;
+		/* No per CPU metadata */
+		return NULL;
+	}
+
+	*ver = ptr[ARM_SPE_HEADER_VERSION];
+	hdr_sz = ptr[ARM_SPE_HEADER_SIZE];
+	*nr_cpu = ptr[ARM_SPE_CPUS_NUM];
+
+	metadata = calloc(*nr_cpu, sizeof(*metadata));
+	if (!metadata)
+		return NULL;
+
+	/* Locate the start address of per CPU metadata */
+	ptr += hdr_sz;
+	per_cpu_sz = (metadata_size - (hdr_sz * sizeof(u64))) / (*nr_cpu);
+
+	for (i = 0; i < *nr_cpu; i++) {
+		metadata[i] = arm_spe__alloc_per_cpu_metadata(ptr, per_cpu_sz);
+		if (!metadata[i])
+			goto err_per_cpu_metadata;
+
+		ptr += per_cpu_sz / sizeof(u64);
+	}
+
+	return metadata;
+
+err_per_cpu_metadata:
+	arm_spe__free_metadata(metadata, *nr_cpu);
+	return NULL;
+}
+
 static void arm_spe_free_queue(void *priv)
 {
 	struct arm_spe_queue *speq = priv;
@@ -1050,6 +1121,7 @@ static void arm_spe_free(struct perf_session *session)
 	auxtrace_heap__free(&spe->heap);
 	arm_spe_free_events(session);
 	session->auxtrace = NULL;
+	arm_spe__free_metadata(spe->metadata, spe->metadata_nr_cpu);
 	free(spe);
 }
 
@@ -1267,15 +1339,26 @@ int arm_spe_process_auxtrace_info(union perf_event *event,
 	const char *cpuid = perf_env__cpuid(session->evlist->env);
 	u64 midr = strtol(cpuid, NULL, 16);
 	struct arm_spe *spe;
-	int err;
+	u64 **metadata = NULL;
+	u64 metadata_ver;
+	int nr_cpu, err;
 
 	if (auxtrace_info->header.size < sizeof(struct perf_record_auxtrace_info) +
 					min_sz)
 		return -EINVAL;
 
+	metadata = arm_spe__alloc_metadata(auxtrace_info, &metadata_ver,
+					   &nr_cpu);
+	if (!metadata && metadata_ver != 1) {
+		pr_err("Failed to parse Arm SPE metadata.\n");
+		return -EINVAL;
+	}
+
 	spe = zalloc(sizeof(struct arm_spe));
-	if (!spe)
-		return -ENOMEM;
+	if (!spe) {
+		err = -ENOMEM;
+		goto err_free_metadata;
+	}
 
 	err = auxtrace_queues__init(&spe->queues);
 	if (err)
@@ -1284,8 +1367,14 @@ int arm_spe_process_auxtrace_info(union perf_event *event,
 	spe->session = session;
 	spe->machine = &session->machines.host; /* No kvm support */
 	spe->auxtrace_type = auxtrace_info->type;
-	spe->pmu_type = auxtrace_info->priv[ARM_SPE_PMU_TYPE];
+	if (metadata_ver == 1)
+		spe->pmu_type = auxtrace_info->priv[ARM_SPE_PMU_TYPE];
+	else
+		spe->pmu_type = auxtrace_info->priv[ARM_SPE_PMU_TYPE_V2];
 	spe->midr = midr;
+	spe->metadata = metadata;
+	spe->metadata_ver = metadata_ver;
+	spe->metadata_nr_cpu = nr_cpu;
 
 	spe->timeless_decoding = arm_spe__is_timeless_decoding(spe);
 
@@ -1346,5 +1435,7 @@ int arm_spe_process_auxtrace_info(union perf_event *event,
 	session->auxtrace = NULL;
 err_free:
 	free(spe);
+err_free_metadata:
+	arm_spe__free_metadata(metadata, nr_cpu);
 	return err;
 }
-- 
2.34.1

[PATCH v4 5/5] perf arm-spe: Dump metadata with version 2

[Leo Yan]

This commit dumps metadata with version 2. It dumps metadata for header
and per CPU data respectively in the arm_spe_print_info() function to
support metadata version 2 format.

After:

  0 0 0x3c0 [0x1b0]: PERF_RECORD_AUXTRACE_INFO type: 4
  Header version     :2
  Header size        :4
  PMU type v2        :13
  CPU number         :8
    Magic            :0x1010101010101010
    CPU #            :0
    Num of params    :3
    MIDR             :0x410fd801
    PMU Type         :-1
    Min Interval     :0
    Magic            :0x1010101010101010
    CPU #            :1
    Num of params    :3
    MIDR             :0x410fd801
    PMU Type         :-1
    Min Interval     :0
    Magic            :0x1010101010101010
    CPU #            :2
    Num of params    :3
    MIDR             :0x410fd870
    PMU Type         :13
    Min Interval     :1024
    Magic            :0x1010101010101010
    CPU #            :3
    Num of params    :3
    MIDR             :0x410fd870
    PMU Type         :13
    Min Interval     :1024
    Magic            :0x1010101010101010
    CPU #            :4
    Num of params    :3
    MIDR             :0x410fd870
    PMU Type         :13
    Min Interval     :1024
    Magic            :0x1010101010101010
    CPU #            :5
    Num of params    :3
    MIDR             :0x410fd870
    PMU Type         :13
    Min Interval     :1024
    Magic            :0x1010101010101010
    CPU #            :6
    Num of params    :3
    MIDR             :0x410fd850
    PMU Type         :-1
    Min Interval     :0
    Magic            :0x1010101010101010
    CPU #            :7
    Num of params    :3
    MIDR             :0x410fd850
    PMU Type         :-1
    Min Interval     :0

Signed-off-by: Leo Yan <leo.yan@arm.com>
Reviewed-by: James Clark <james.clark@linaro.org>
---
 tools/perf/util/arm-spe.c | 54 +++++++++++++++++++++++++++++++++++----
 1 file changed, 49 insertions(+), 5 deletions(-)

diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
index 8b2af1c7dc9b..13fd2c8afebd 100644
--- a/tools/perf/util/arm-spe.c
+++ b/tools/perf/util/arm-spe.c
@@ -1133,16 +1133,60 @@ static bool arm_spe_evsel_is_auxtrace(struct perf_session *session,
 	return evsel->core.attr.type == spe->pmu_type;
 }
 
-static const char * const arm_spe_info_fmts[] = {
-	[ARM_SPE_PMU_TYPE]		= "  PMU Type           %"PRId64"\n",
+static const char * const metadata_hdr_v1_fmts[] = {
+	[ARM_SPE_PMU_TYPE]		= "  PMU Type           :%"PRId64"\n",
+	[ARM_SPE_PER_CPU_MMAPS]		= "  Per CPU mmaps      :%"PRId64"\n",
 };
 
-static void arm_spe_print_info(__u64 *arr)
+static const char * const metadata_hdr_fmts[] = {
+	[ARM_SPE_HEADER_VERSION]	= "  Header version     :%"PRId64"\n",
+	[ARM_SPE_HEADER_SIZE]		= "  Header size        :%"PRId64"\n",
+	[ARM_SPE_PMU_TYPE_V2]		= "  PMU type v2        :%"PRId64"\n",
+	[ARM_SPE_CPUS_NUM]		= "  CPU number         :%"PRId64"\n",
+};
+
+static const char * const metadata_per_cpu_fmts[] = {
+	[ARM_SPE_MAGIC]			= "    Magic            :0x%"PRIx64"\n",
+	[ARM_SPE_CPU]			= "    CPU #            :%"PRId64"\n",
+	[ARM_SPE_CPU_NR_PARAMS]		= "    Num of params    :%"PRId64"\n",
+	[ARM_SPE_CPU_MIDR]		= "    MIDR             :0x%"PRIx64"\n",
+	[ARM_SPE_CPU_PMU_TYPE]		= "    PMU Type         :%"PRId64"\n",
+	[ARM_SPE_CAP_MIN_IVAL]		= "    Min Interval     :%"PRId64"\n",
+};
+
+static void arm_spe_print_info(struct arm_spe *spe, __u64 *arr)
 {
+	unsigned int i, cpu, hdr_size, cpu_num, cpu_size;
+	const char * const *hdr_fmts;
+
 	if (!dump_trace)
 		return;
 
-	fprintf(stdout, arm_spe_info_fmts[ARM_SPE_PMU_TYPE], arr[ARM_SPE_PMU_TYPE]);
+	if (spe->metadata_ver == 1) {
+		cpu_num = 0;
+		hdr_size = ARM_SPE_AUXTRACE_V1_PRIV_MAX;
+		hdr_fmts = metadata_hdr_v1_fmts;
+	} else {
+		cpu_num = arr[ARM_SPE_CPUS_NUM];
+		hdr_size = arr[ARM_SPE_HEADER_SIZE];
+		hdr_fmts = metadata_hdr_fmts;
+	}
+
+	for (i = 0; i < hdr_size; i++)
+		fprintf(stdout, hdr_fmts[i], arr[i]);
+
+	arr += hdr_size;
+	for (cpu = 0; cpu < cpu_num; cpu++) {
+		/*
+		 * The parameters from ARM_SPE_MAGIC to ARM_SPE_CPU_NR_PARAMS
+		 * are fixed. The sequential parameter size is decided by the
+		 * field 'ARM_SPE_CPU_NR_PARAMS'.
+		 */
+		cpu_size = (ARM_SPE_CPU_NR_PARAMS + 1) + arr[ARM_SPE_CPU_NR_PARAMS];
+		for (i = 0; i < cpu_size; i++)
+			fprintf(stdout, metadata_per_cpu_fmts[i], arr[i]);
+		arr += cpu_size;
+	}
 }
 
 static void arm_spe_set_event_name(struct evlist *evlist, u64 id,
@@ -1407,7 +1451,7 @@ int arm_spe_process_auxtrace_info(union perf_event *event,
 	spe->auxtrace.evsel_is_auxtrace = arm_spe_evsel_is_auxtrace;
 	session->auxtrace = &spe->auxtrace;
 
-	arm_spe_print_info(&auxtrace_info->priv[0]);
+	arm_spe_print_info(spe, &auxtrace_info->priv[0]);
 
 	if (dump_trace)
 		return 0;
-- 
2.34.1

Re: [PATCH v4 0/5] perf arm-spe: Introduce metadata version 2

[Namhyung Kim]

On Thu, 03 Oct 2024 19:42:57 +0100, Leo Yan wrote:

> This patch series enhances Arm SPE metadata in the Perf file to a
> version 2 format and maintains backward compatibility for metadata v1.
> 
> The version 2 metadata contains a global structure with fields for
> metadata header version number, header size, CPU numbers. It also
> expands to store per-CPU metadata, which includes the CPU logical
> number in the kernel, MIDR, and capacity values associated with Arm SPE.
> 
> [...]

Applied to perf-tools-next, thanks!

Best regards,
Namhyung