Changes since v2:
- Refact cmn identifier and use model and revision to form identifier.
- Let "Compat" support matching multiple identifier.
- Improved the ali_drw PMU event alias Brief Description.
- Update ali_drw PMU metric usage in documentation.
Changes since RFC:
- Refact arm-cmn PMU identifier.
- Not add arm-cmn PMU aliasing currently because it's Eventcode is
difficult to define.
- Rename ali_drw PMU identifier and Unit name.
- Divide ali_drw PMU metric and aliasing into two patches.
Add an identifier sysfs file for the yitian710 SoC DDR and arm CMN to
allow userspace to identify the specific implementation of the device,
so that the perf tool can match the corresponding uncore events and
metrics through the identifier. Then added several general CMN metrics
and yitian710 soc DDR metrics and events alias.
$perf list:
...
ali_drw:
chi_rxdat
[A packet at CHI RXDAT interface (write data). Unit: ali_drw]
chi_rxrsp
[A packet at CHI RXRSP interface. Unit: ali_drw]
chi_txdat
[A packet at CHI TXDAT interface (read data). Unit: ali_drw]
chi_txreq
[A packet at CHI TXREQ interface (request). Unit: ali_drw]
cycle
[The ddr cycle. Unit: ali_drw]
...
arm_cmn:
mc_message_retry_rate
[The memory controller request retries rate indicates whether the memory controller is the bottleneck. Unit: arm_cmn ]
rni_actual_read_bandwidth.all
[This event measure the actual bandwidth(MB/sec) that RN-I bridge sends to the interconnect. Unit: arm_cmn ]
rni_actual_write_bandwidth.all
[This event measures the actual write bandwidth(MB/sec) at RN-I bridges. Unit: arm_cmn ]
rni_retry_rate
[RN-I bridge retry rate indicates whether the memory controller is the bottleneck. Unit: arm_cmn ]
sbsx_actual_write_bandwidth.all
[sbsx actual write bandwidth(MB/sec). Unit: arm_cmn ]
sf_hit_rate
[Snoop filter hit rate can be used to measure the Snoop Filter efficiency. Unit: arm_cmn ]
slc_miss_rate
[The system level cache miss rate include. Unit: arm_cmn ]
ali_drw:
ddr_read_bandwidth.all
[The ddr read bandwidth(MB/s). Unit: ali_drw ]
ddr_write_bandwidth.all
[The ddr write bandwidth(MB/s). Unit: ali_drw ]
...
$perf stat -M ddr_read_bandwidth.all ./test
Performance counter stats for 'system wide':
38,150 hif_rd # 2.4 MB/s ddr_read_bandwidth.all
1,000,957,941 ns duration_time
1.000957941 seconds time elapsed
Jing Zhang (7):
driver/perf: Add identifier sysfs file for CMN
perf metric: Event "Compat" value supports matching multiple
identifiers
perf vendor events: Add JSON metrics for CMN
driver/perf: Add identifier sysfs file for Yitian 710 DDR
perf jevents: Add support for Yitian 710 DDR PMU aliasing
perf vendor events: Add JSON metrics for Yitian 710 DDR
docs: perf: Update metric usage for Alibaba's T-Head PMU driver
Documentation/admin-guide/perf/alibaba_pmu.rst | 5 +
drivers/perf/alibaba_uncore_drw_pmu.c | 27 ++
drivers/perf/arm-cmn.c | 79 ++++-
.../pmu-events/arch/arm64/arm/cmn/sys/metrics.json | 74 ++++
.../arm64/freescale/yitian710/sys/ali_drw.json | 373 +++++++++++++++++++++
.../arm64/freescale/yitian710/sys/metrics.json | 20 ++
tools/perf/pmu-events/jevents.py | 2 +
tools/perf/util/metricgroup.c | 24 +-
8 files changed, 595 insertions(+), 9 deletions(-)
create mode 100644 tools/perf/pmu-events/arch/arm64/arm/cmn/sys/metrics.json
create mode 100644 tools/perf/pmu-events/arch/arm64/freescale/yitian710/sys/ali_drw.json
create mode 100644 tools/perf/pmu-events/arch/arm64/freescale/yitian710/sys/metrics.json
--
1.8.3.1
Add JSON metrics for T-HEAD Yitian 710 SoC DDR.
Signed-off-by: Jing Zhang <[email protected]>
Acked-by: Ian Rogers <[email protected]>
Reviewed-by: John Garry <[email protected]>
---
.../arch/arm64/freescale/yitian710/sys/metrics.json | 20 ++++++++++++++++++++
1 file changed, 20 insertions(+)
create mode 100644 tools/perf/pmu-events/arch/arm64/freescale/yitian710/sys/metrics.json
diff --git a/tools/perf/pmu-events/arch/arm64/freescale/yitian710/sys/metrics.json b/tools/perf/pmu-events/arch/arm64/freescale/yitian710/sys/metrics.json
new file mode 100644
index 0000000..1a92477
--- /dev/null
+++ b/tools/perf/pmu-events/arch/arm64/freescale/yitian710/sys/metrics.json
@@ -0,0 +1,20 @@
+[
+ {
+ "MetricName": "ddr_read_bandwidth.all",
+ "BriefDescription": "The ddr read bandwidth(MB/s).",
+ "MetricGroup": "ali_drw",
+ "MetricExpr": "hif_rd * 64 / 1e6 / duration_time",
+ "ScaleUnit": "1MB/s",
+ "Unit": "ali_drw",
+ "Compat": "ali_drw_pmu"
+ },
+ {
+ "MetricName": "ddr_write_bandwidth.all",
+ "BriefDescription": "The ddr write bandwidth(MB/s).",
+ "MetricGroup": "ali_drw",
+ "MetricExpr": "(hif_wr + hif_rmw) * 64 / 1e6 / duration_time",
+ "ScaleUnit": "1MB/s",
+ "Unit": "ali_drw",
+ "Compat": "ali_drw_pmu"
+ }
+]
--
1.8.3.1
To allow userspace to identify the specific implementation of the device,
add an "identifier" sysfs file.
The perf tool can match the Yitian 710 DDR metric through the identifier.
Signed-off-by: Jing Zhang <[email protected]>
Acked-by: Ian Rogers <[email protected]>
Reviewed-by: Shuai Xue <[email protected]>
---
drivers/perf/alibaba_uncore_drw_pmu.c | 27 +++++++++++++++++++++++++++
1 file changed, 27 insertions(+)
diff --git a/drivers/perf/alibaba_uncore_drw_pmu.c b/drivers/perf/alibaba_uncore_drw_pmu.c
index a7689fe..fe075fd 100644
--- a/drivers/perf/alibaba_uncore_drw_pmu.c
+++ b/drivers/perf/alibaba_uncore_drw_pmu.c
@@ -236,10 +236,37 @@ static ssize_t ali_drw_pmu_cpumask_show(struct device *dev,
.attrs = ali_drw_pmu_cpumask_attrs,
};
+static ssize_t ali_drw_pmu_identifier_show(struct device *dev,
+ struct device_attribute *attr,
+ char *page)
+{
+ return sysfs_emit(page, "%s\n", "ali_drw_pmu");
+}
+
+static umode_t ali_drw_pmu_identifier_attr_visible(struct kobject *kobj,
+ struct attribute *attr, int n)
+{
+ return attr->mode;
+}
+
+static struct device_attribute ali_drw_pmu_identifier_attr =
+ __ATTR(identifier, 0444, ali_drw_pmu_identifier_show, NULL);
+
+static struct attribute *ali_drw_pmu_identifier_attrs[] = {
+ &ali_drw_pmu_identifier_attr.attr,
+ NULL
+};
+
+static const struct attribute_group ali_drw_pmu_identifier_attr_group = {
+ .attrs = ali_drw_pmu_identifier_attrs,
+ .is_visible = ali_drw_pmu_identifier_attr_visible
+};
+
static const struct attribute_group *ali_drw_pmu_attr_groups[] = {
&ali_drw_pmu_events_attr_group,
&ali_drw_pmu_cpumask_attr_group,
&ali_drw_pmu_format_group,
+ &ali_drw_pmu_identifier_attr_group,
NULL,
};
--
1.8.3.1
Alibaba's T-Head ali_drw PMU supports DDR bandwidth metrics. Update
its usage in the documentation.
Signed-off-by: Jing Zhang <[email protected]>
---
Documentation/admin-guide/perf/alibaba_pmu.rst | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/Documentation/admin-guide/perf/alibaba_pmu.rst b/Documentation/admin-guide/perf/alibaba_pmu.rst
index 11de998..7d84002 100644
--- a/Documentation/admin-guide/perf/alibaba_pmu.rst
+++ b/Documentation/admin-guide/perf/alibaba_pmu.rst
@@ -88,6 +88,11 @@ data bandwidth::
-e ali_drw_27080/hif_rmw/ \
-e ali_drw_27080/cycle/ -- sleep 10
+Example usage of counting all memory read/write bandwidth by metric::
+
+ perf stat -M ddr_read_bandwidth.all -- sleep 10
+ perf stat -M ddr_write_bandwidth.all -- sleep 10
+
The average DRAM bandwidth can be calculated as follows:
- Read Bandwidth = perf_hif_rd * DDRC_WIDTH * DDRC_Freq / DDRC_Cycle
--
1.8.3.1
On Tue, May 30, 2023 at 2:19 AM Jing Zhang <[email protected]> wrote:
>
> Alibaba's T-Head ali_drw PMU supports DDR bandwidth metrics. Update
> its usage in the documentation.
>
> Signed-off-by: Jing Zhang <[email protected]>
Acked-by: Ian Rogers <[email protected]>
Thanks,
Ian
> ---
> Documentation/admin-guide/perf/alibaba_pmu.rst | 5 +++++
> 1 file changed, 5 insertions(+)
>
> diff --git a/Documentation/admin-guide/perf/alibaba_pmu.rst b/Documentation/admin-guide/perf/alibaba_pmu.rst
> index 11de998..7d84002 100644
> --- a/Documentation/admin-guide/perf/alibaba_pmu.rst
> +++ b/Documentation/admin-guide/perf/alibaba_pmu.rst
> @@ -88,6 +88,11 @@ data bandwidth::
> -e ali_drw_27080/hif_rmw/ \
> -e ali_drw_27080/cycle/ -- sleep 10
>
> +Example usage of counting all memory read/write bandwidth by metric::
> +
> + perf stat -M ddr_read_bandwidth.all -- sleep 10
> + perf stat -M ddr_write_bandwidth.all -- sleep 10
> +
> The average DRAM bandwidth can be calculated as follows:
>
> - Read Bandwidth = perf_hif_rd * DDRC_WIDTH * DDRC_Freq / DDRC_Cycle
> --
> 1.8.3.1
>