[PATCH v3 1/2] perf arm-spe: Prepare for adding data source packet implementations for other cores

Ilkka Koskinen posted 2 patches 2 weeks, 3 days ago
There is a newer version of this series
[PATCH v3 1/2] perf arm-spe: Prepare for adding data source packet implementations for other cores
Posted by Ilkka Koskinen 2 weeks, 3 days ago
Split Data Source Packet handling to prepare adding support for
other implementations.

Signed-off-by: Ilkka Koskinen <ilkka@os.amperecomputing.com>
---
 tools/perf/util/arm-spe.c | 65 ++++++++++++++++++++++++---------------
 1 file changed, 40 insertions(+), 25 deletions(-)

diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
index dbf13f47879c..b222557cc27a 100644
--- a/tools/perf/util/arm-spe.c
+++ b/tools/perf/util/arm-spe.c
@@ -103,6 +103,18 @@ struct arm_spe_queue {
 	u32				flags;
 };
 
+struct data_src {
+	struct midr_range midr_range;
+	void (*ds_synth)(const struct arm_spe_record *record,
+			 union perf_mem_data_src *data_src);
+};
+
+#define DS(range, func)			\
+	{						\
+		.midr_range = range,			\
+		.ds_synth = arm_spe__synth_##func,	\
+	}
+
 static void arm_spe_dump(struct arm_spe *spe __maybe_unused,
 			 unsigned char *buf, size_t len)
 {
@@ -430,19 +442,6 @@ static int arm_spe__synth_instruction_sample(struct arm_spe_queue *speq,
 	return arm_spe_deliver_synth_event(spe, speq, event, &sample);
 }
 
-static const struct midr_range common_ds_encoding_cpus[] = {
-	MIDR_ALL_VERSIONS(MIDR_CORTEX_A720),
-	MIDR_ALL_VERSIONS(MIDR_CORTEX_A725),
-	MIDR_ALL_VERSIONS(MIDR_CORTEX_X1C),
-	MIDR_ALL_VERSIONS(MIDR_CORTEX_X3),
-	MIDR_ALL_VERSIONS(MIDR_CORTEX_X925),
-	MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N1),
-	MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N2),
-	MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V1),
-	MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V2),
-	{},
-};
-
 static void arm_spe__sample_flags(struct arm_spe_queue *speq)
 {
 	const struct arm_spe_record *record = &speq->decoder->record;
@@ -532,6 +531,19 @@ static void arm_spe__synth_data_source_common(const struct arm_spe_record *recor
 	}
 }
 
+static const struct data_src data_sources[] = {
+	DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_A720), data_source_common),
+	DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_A725), data_source_common),
+	DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_X1C), data_source_common),
+	DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_X3), data_source_common),
+	DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_X925), data_source_common),
+	DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N1), data_source_common),
+	DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N2), data_source_common),
+	DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V1), data_source_common),
+	DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V2), data_source_common),
+	{},
+};
+
 static void arm_spe__synth_memory_level(const struct arm_spe_record *record,
 					union perf_mem_data_src *data_src)
 {
@@ -555,12 +567,14 @@ static void arm_spe__synth_memory_level(const struct arm_spe_record *record,
 		data_src->mem_lvl |= PERF_MEM_LVL_REM_CCE1;
 }
 
-static bool arm_spe__is_common_ds_encoding(struct arm_spe_queue *speq)
+static bool arm_spe__synth_ds(struct arm_spe_queue *speq,
+			      const struct arm_spe_record *record,
+			      union perf_mem_data_src *data_src)
 {
 	struct arm_spe *spe = speq->spe;
-	bool is_in_cpu_list;
+	const struct data_src *src = data_sources;
 	u64 *metadata = NULL;
-	u64 midr = 0;
+	u64 midr;
 
 	/* Metadata version 1 assumes all CPUs are the same (old behavior) */
 	if (spe->metadata_ver == 1) {
@@ -592,18 +606,21 @@ static bool arm_spe__is_common_ds_encoding(struct arm_spe_queue *speq)
 		midr = metadata[ARM_SPE_CPU_MIDR];
 	}
 
-	is_in_cpu_list = is_midr_in_range_list(midr, common_ds_encoding_cpus);
-	if (is_in_cpu_list)
-		return true;
-	else
-		return false;
+	while (src->midr_range.model) {
+		if (is_midr_in_range(midr, &src->midr_range)) {
+			src->ds_synth(record, data_src);
+			return true;
+		}
+		src++;
+	}
+
+	return false;
 }
 
 static u64 arm_spe__synth_data_source(struct arm_spe_queue *speq,
 				      const struct arm_spe_record *record)
 {
 	union perf_mem_data_src	data_src = { .mem_op = PERF_MEM_OP_NA };
-	bool is_common = arm_spe__is_common_ds_encoding(speq);
 
 	if (record->op & ARM_SPE_OP_LD)
 		data_src.mem_op = PERF_MEM_OP_LOAD;
@@ -612,9 +629,7 @@ static u64 arm_spe__synth_data_source(struct arm_spe_queue *speq,
 	else
 		return 0;
 
-	if (is_common)
-		arm_spe__synth_data_source_common(record, &data_src);
-	else
+	if (!arm_spe__synth_ds(speq, record, &data_src))
 		arm_spe__synth_memory_level(record, &data_src);
 
 	if (record->type & (ARM_SPE_TLB_ACCESS | ARM_SPE_TLB_MISS)) {
-- 
2.47.0
Re: [PATCH v3 1/2] perf arm-spe: Prepare for adding data source packet implementations for other cores
Posted by Leo Yan 2 weeks, 2 days ago
Hi Ilkka,

This is a good refactoring for me.  Just several minor comments.

On Wed, Nov 06, 2024 at 07:37:39PM +0000, Ilkka Koskinen wrote:
> 
> Split Data Source Packet handling to prepare adding support for
> other implementations.
> 
> Signed-off-by: Ilkka Koskinen <ilkka@os.amperecomputing.com>
> ---
>  tools/perf/util/arm-spe.c | 65 ++++++++++++++++++++++++---------------
>  1 file changed, 40 insertions(+), 25 deletions(-)
> 
> diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
> index dbf13f47879c..b222557cc27a 100644
> --- a/tools/perf/util/arm-spe.c
> +++ b/tools/perf/util/arm-spe.c
> @@ -103,6 +103,18 @@ struct arm_spe_queue {
>         u32                             flags;
>  };
> 
> +struct data_src {
> +       struct midr_range midr_range;
> +       void (*ds_synth)(const struct arm_spe_record *record,
> +                        union perf_mem_data_src *data_src);
> +};

The naming is a bit mess. The data structure and the parameter both
are called "data_src", though this will not cause building issue.

How about rename the structure "data_src" to "data_source_handle" or
"data_source_class"?

For the "midr_range" field, I'd like to change it to a pointer:

  struct midr_range *midr_range;

Please see below comments, which will present the reason for defining
it as a pointer.

> +
> +#define DS(range, func)                        \
> +       {                                               \
> +               .midr_range = range,                    \
> +               .ds_synth = arm_spe__synth_##func,      \
> +       }
> +
>  static void arm_spe_dump(struct arm_spe *spe __maybe_unused,
>                          unsigned char *buf, size_t len)
>  {
> @@ -430,19 +442,6 @@ static int arm_spe__synth_instruction_sample(struct arm_spe_queue *speq,
>         return arm_spe_deliver_synth_event(spe, speq, event, &sample);
>  }
> 
> -static const struct midr_range common_ds_encoding_cpus[] = {
> -       MIDR_ALL_VERSIONS(MIDR_CORTEX_A720),
> -       MIDR_ALL_VERSIONS(MIDR_CORTEX_A725),
> -       MIDR_ALL_VERSIONS(MIDR_CORTEX_X1C),
> -       MIDR_ALL_VERSIONS(MIDR_CORTEX_X3),
> -       MIDR_ALL_VERSIONS(MIDR_CORTEX_X925),
> -       MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N1),
> -       MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N2),
> -       MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V1),
> -       MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V2),
> -       {},
> -};

We can keep this data structure.  For Ampere CPUs, you can add a new
data structure:

  static const struct midr_range ampereone_ds_encoding_cpus[] = {
      MIDR_ALL_VERSIONS(MIDR_AMPERE1A),
      {},
  };

> -
>  static void arm_spe__sample_flags(struct arm_spe_queue *speq)
>  {
>         const struct arm_spe_record *record = &speq->decoder->record;
> @@ -532,6 +531,19 @@ static void arm_spe__synth_data_source_common(const struct arm_spe_record *recor
>         }
>  }
> 
> +static const struct data_src data_sources[] = {
> +       DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_A720), data_source_common),
> +       DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_A725), data_source_common),
> +       DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_X1C), data_source_common),
> +       DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_X3), data_source_common),
> +       DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_X925), data_source_common),
> +       DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N1), data_source_common),
> +       DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N2), data_source_common),
> +       DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V1), data_source_common),
> +       DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V2), data_source_common),
> +       {},
> +};
> +

As a result, we can simplify the structure as:

  static const struct data_src data_sources[] = {
         DS(common_ds_encoding_cpus, data_source_common),
         DS(ampereone_ds_encoding_cpus, data_source_ampereone),
  };

>  static void arm_spe__synth_memory_level(const struct arm_spe_record *record,
>                                         union perf_mem_data_src *data_src)
>  {
> @@ -555,12 +567,14 @@ static void arm_spe__synth_memory_level(const struct arm_spe_record *record,
>                 data_src->mem_lvl |= PERF_MEM_LVL_REM_CCE1;
>  }
> 
> -static bool arm_spe__is_common_ds_encoding(struct arm_spe_queue *speq)
> +static bool arm_spe__synth_ds(struct arm_spe_queue *speq,
> +                             const struct arm_spe_record *record,
> +                             union perf_mem_data_src *data_src)
>  {
>         struct arm_spe *spe = speq->spe;
> -       bool is_in_cpu_list;
> +       const struct data_src *src = data_sources;
>         u64 *metadata = NULL;
> -       u64 midr = 0;
> +       u64 midr;
> 
>         /* Metadata version 1 assumes all CPUs are the same (old behavior) */
>         if (spe->metadata_ver == 1) {
> @@ -592,18 +606,21 @@ static bool arm_spe__is_common_ds_encoding(struct arm_spe_queue *speq)
>                 midr = metadata[ARM_SPE_CPU_MIDR];
>         }
> 
> -       is_in_cpu_list = is_midr_in_range_list(midr, common_ds_encoding_cpus);
> -       if (is_in_cpu_list)
> -               return true;
> -       else
> -               return false;
> +       while (src->midr_range.model) {
> +               if (is_midr_in_range(midr, &src->midr_range)) {
> +                       src->ds_synth(record, data_src);
> +                       return true;
> +               }
> +               src++;
> +       }

Here we can traverse the 'data_sources' array:

          for (i = 0; i < ARRAY_SIZE(data_sources); i++) {
               if (is_midr_in_range(midr, data_sources[i]->midr_range)) {
                   ...
               }
          }

Thanks,
Leo

> +
> +       return false;
>  }
> 
>  static u64 arm_spe__synth_data_source(struct arm_spe_queue *speq,
>                                       const struct arm_spe_record *record)
>  {
>         union perf_mem_data_src data_src = { .mem_op = PERF_MEM_OP_NA };
> -       bool is_common = arm_spe__is_common_ds_encoding(speq);
> 
>         if (record->op & ARM_SPE_OP_LD)
>                 data_src.mem_op = PERF_MEM_OP_LOAD;
> @@ -612,9 +629,7 @@ static u64 arm_spe__synth_data_source(struct arm_spe_queue *speq,
>         else
>                 return 0;
> 
> -       if (is_common)
> -               arm_spe__synth_data_source_common(record, &data_src);
> -       else
> +       if (!arm_spe__synth_ds(speq, record, &data_src))
>                 arm_spe__synth_memory_level(record, &data_src);
> 
>         if (record->type & (ARM_SPE_TLB_ACCESS | ARM_SPE_TLB_MISS)) {
> --
> 2.47.0
> 
>
Re: [PATCH v3 1/2] perf arm-spe: Prepare for adding data source packet implementations for other cores
Posted by Ilkka Koskinen 2 weeks, 2 days ago
Hi Leo,

On Thu, 7 Nov 2024, Leo Yan wrote:
> Hi Ilkka,
>
> This is a good refactoring for me.  Just several minor comments.
>
> On Wed, Nov 06, 2024 at 07:37:39PM +0000, Ilkka Koskinen wrote:
>>
>> Split Data Source Packet handling to prepare adding support for
>> other implementations.
>>
>> Signed-off-by: Ilkka Koskinen <ilkka@os.amperecomputing.com>
>> ---
>>  tools/perf/util/arm-spe.c | 65 ++++++++++++++++++++++++---------------
>>  1 file changed, 40 insertions(+), 25 deletions(-)
>>
>> diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
>> index dbf13f47879c..b222557cc27a 100644
>> --- a/tools/perf/util/arm-spe.c
>> +++ b/tools/perf/util/arm-spe.c
>> @@ -103,6 +103,18 @@ struct arm_spe_queue {
>>         u32                             flags;
>>  };
>>
>> +struct data_src {
>> +       struct midr_range midr_range;
>> +       void (*ds_synth)(const struct arm_spe_record *record,
>> +                        union perf_mem_data_src *data_src);
>> +};
>
> The naming is a bit mess. The data structure and the parameter both
> are called "data_src", though this will not cause building issue.
>
> How about rename the structure "data_src" to "data_source_handle" or
> "data_source_class"?

Yeah, I forgot to revisit the naming part. I like "data"source_handle",
that should clarify it quite a bit.


>
> For the "midr_range" field, I'd like to change it to a pointer:
>
>  struct midr_range *midr_range;
>
> Please see below comments, which will present the reason for defining
> it as a pointer.
>
>> +
>> +#define DS(range, func)                        \
>> +       {                                               \
>> +               .midr_range = range,                    \
>> +               .ds_synth = arm_spe__synth_##func,      \
>> +       }
>> +
>>  static void arm_spe_dump(struct arm_spe *spe __maybe_unused,
>>                          unsigned char *buf, size_t len)
>>  {
>> @@ -430,19 +442,6 @@ static int arm_spe__synth_instruction_sample(struct arm_spe_queue *speq,
>>         return arm_spe_deliver_synth_event(spe, speq, event, &sample);
>>  }
>>
>> -static const struct midr_range common_ds_encoding_cpus[] = {
>> -       MIDR_ALL_VERSIONS(MIDR_CORTEX_A720),
>> -       MIDR_ALL_VERSIONS(MIDR_CORTEX_A725),
>> -       MIDR_ALL_VERSIONS(MIDR_CORTEX_X1C),
>> -       MIDR_ALL_VERSIONS(MIDR_CORTEX_X3),
>> -       MIDR_ALL_VERSIONS(MIDR_CORTEX_X925),
>> -       MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N1),
>> -       MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N2),
>> -       MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V1),
>> -       MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V2),
>> -       {},
>> -};
>
> We can keep this data structure.  For Ampere CPUs, you can add a new
> data structure:
>
>  static const struct midr_range ampereone_ds_encoding_cpus[] = {
>      MIDR_ALL_VERSIONS(MIDR_AMPERE1A),
>      {},
>  };

Sounds good to me. I change all those.

Cheers, Ilkka


>
>> -
>>  static void arm_spe__sample_flags(struct arm_spe_queue *speq)
>>  {
>>         const struct arm_spe_record *record = &speq->decoder->record;
>> @@ -532,6 +531,19 @@ static void arm_spe__synth_data_source_common(const struct arm_spe_record *recor
>>         }
>>  }
>>
>> +static const struct data_src data_sources[] = {
>> +       DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_A720), data_source_common),
>> +       DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_A725), data_source_common),
>> +       DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_X1C), data_source_common),
>> +       DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_X3), data_source_common),
>> +       DS(MIDR_ALL_VERSIONS(MIDR_CORTEX_X925), data_source_common),
>> +       DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N1), data_source_common),
>> +       DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_N2), data_source_common),
>> +       DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V1), data_source_common),
>> +       DS(MIDR_ALL_VERSIONS(MIDR_NEOVERSE_V2), data_source_common),
>> +       {},
>> +};
>> +
>
> As a result, we can simplify the structure as:
>
>  static const struct data_src data_sources[] = {
>         DS(common_ds_encoding_cpus, data_source_common),
>         DS(ampereone_ds_encoding_cpus, data_source_ampereone),
>  };
>
>>  static void arm_spe__synth_memory_level(const struct arm_spe_record *record,
>>                                         union perf_mem_data_src *data_src)
>>  {
>> @@ -555,12 +567,14 @@ static void arm_spe__synth_memory_level(const struct arm_spe_record *record,
>>                 data_src->mem_lvl |= PERF_MEM_LVL_REM_CCE1;
>>  }
>>
>> -static bool arm_spe__is_common_ds_encoding(struct arm_spe_queue *speq)
>> +static bool arm_spe__synth_ds(struct arm_spe_queue *speq,
>> +                             const struct arm_spe_record *record,
>> +                             union perf_mem_data_src *data_src)
>>  {
>>         struct arm_spe *spe = speq->spe;
>> -       bool is_in_cpu_list;
>> +       const struct data_src *src = data_sources;
>>         u64 *metadata = NULL;
>> -       u64 midr = 0;
>> +       u64 midr;
>>
>>         /* Metadata version 1 assumes all CPUs are the same (old behavior) */
>>         if (spe->metadata_ver == 1) {
>> @@ -592,18 +606,21 @@ static bool arm_spe__is_common_ds_encoding(struct arm_spe_queue *speq)
>>                 midr = metadata[ARM_SPE_CPU_MIDR];
>>         }
>>
>> -       is_in_cpu_list = is_midr_in_range_list(midr, common_ds_encoding_cpus);
>> -       if (is_in_cpu_list)
>> -               return true;
>> -       else
>> -               return false;
>> +       while (src->midr_range.model) {
>> +               if (is_midr_in_range(midr, &src->midr_range)) {
>> +                       src->ds_synth(record, data_src);
>> +                       return true;
>> +               }
>> +               src++;
>> +       }
>
> Here we can traverse the 'data_sources' array:
>
>          for (i = 0; i < ARRAY_SIZE(data_sources); i++) {
>               if (is_midr_in_range(midr, data_sources[i]->midr_range)) {
>                   ...
>               }
>          }
>
> Thanks,
> Leo
>
>> +
>> +       return false;
>>  }
>>
>>  static u64 arm_spe__synth_data_source(struct arm_spe_queue *speq,
>>                                       const struct arm_spe_record *record)
>>  {
>>         union perf_mem_data_src data_src = { .mem_op = PERF_MEM_OP_NA };
>> -       bool is_common = arm_spe__is_common_ds_encoding(speq);
>>
>>         if (record->op & ARM_SPE_OP_LD)
>>                 data_src.mem_op = PERF_MEM_OP_LOAD;
>> @@ -612,9 +629,7 @@ static u64 arm_spe__synth_data_source(struct arm_spe_queue *speq,
>>         else
>>                 return 0;
>>
>> -       if (is_common)
>> -               arm_spe__synth_data_source_common(record, &data_src);
>> -       else
>> +       if (!arm_spe__synth_ds(speq, record, &data_src))
>>                 arm_spe__synth_memory_level(record, &data_src);
>>
>>         if (record->type & (ARM_SPE_TLB_ACCESS | ARM_SPE_TLB_MISS)) {
>> --
>> 2.47.0
>>
>>
>