[PATCH V5 0/3] perf: imx8_ddr_perf: add AXI ID filter

[PATCH V5 0/3] perf: imx8_ddr_perf: add AXI ID filter

[Joakim Zhang]

Add AXI ID filter for imx8m ddr perf.

Joakim Zhang (3):
  perf: imx8_ddr_perf: add AXI ID filter support
  Documentation: admin-guide: perf: add i.MX8 ddr pmu user doc
  MAINTAINERS: add imx8 ddr perf admin-guide maintainer information

 Documentation/admin-guide/perf/imx-ddr.rst | 30 +++++++++++
 MAINTAINERS                                |  1 +
 drivers/perf/fsl_imx8_ddr_perf.c           | 63 +++++++++++++++++++++-
 3 files changed, 92 insertions(+), 2 deletions(-)
 create mode 100644 Documentation/admin-guide/perf/imx-ddr.rst

-- 
2.17.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

[PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter support

[Joakim Zhang]

AXI filtering is used by CSV modes 0x41 and 0x42 to count reads or
writes with an ARID or AXID matching filter setting. Granularity is at
subsystem level. Implementation does not allow filtring between masters
within a subsystem. Filter is defined with 2 configuration registers.

--AXI_ID defines AxID matching value
--AXI_MASKING defines which bits of AxID are meaningful for the matching

When non-masked bits are matching corresponding AXI_ID bits then counter
is incremented. This filter allows counting read or write access from a
subsystem or multiple subsystems.

Perf counter is incremented if AxID && AXI_MASKING == AXI_ID && AXI_MASKING

AXI_ID and AXI_MASKING are mapped on DPCR1 register in performance counter.

Read and write AXI ID filter should write same value to DPCR1 if want to
specify at the same time as this filter is shared between counters.

e.g.
perf stat -a -e imx8_ddr0/axi-id-read,axi_id=0xMMMMDDDD/,imx8_ddr0/axi-id-write,axi_id=0xMMMMDDDD/ cmd
MMMM: AXI_MASKING
DDDD: AXI_ID

ChangeLog:
V1 -> V2:
	* add error log if user specifies read/write AXI ID filter at
	the same time.
	* of_device_get_match_data() instead of of_match_device(), and
	remove the check of return value.
V2 -> V3:
	* move the AXI ID check to event_add().
	* add support for same value of axi_id.
V3 -> V4:
	* move the AXI ID check to event_init().
V4 -> V5:
	* reject event group if AXI ID not consistent in event_init().

Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
---
 drivers/perf/fsl_imx8_ddr_perf.c | 63 +++++++++++++++++++++++++++++++-
 1 file changed, 61 insertions(+), 2 deletions(-)

diff --git a/drivers/perf/fsl_imx8_ddr_perf.c b/drivers/perf/fsl_imx8_ddr_perf.c
index 63fe21600072..f25cf5cbe156 100644
--- a/drivers/perf/fsl_imx8_ddr_perf.c
+++ b/drivers/perf/fsl_imx8_ddr_perf.c
@@ -42,9 +42,22 @@
 
 static DEFINE_IDA(ddr_ida);
 
+/* DDR Perf hardware feature */
+#define DDR_CAP_AXI_ID_FILTER		0x1	/* support AXI ID filter */
+
+struct fsl_ddr_devtype_data {
+	unsigned int quirks;	/* quirks needed for different DDR Perf core */
+};
+
+static const struct fsl_ddr_devtype_data imx8_devtype_data;
+
+static const struct fsl_ddr_devtype_data imx8m_devtype_data = {
+	.quirks = DDR_CAP_AXI_ID_FILTER,
+};
+
 static const struct of_device_id imx_ddr_pmu_dt_ids[] = {
-	{ .compatible = "fsl,imx8-ddr-pmu",},
-	{ .compatible = "fsl,imx8m-ddr-pmu",},
+	{ .compatible = "fsl,imx8-ddr-pmu", .data = &imx8_devtype_data},
+	{ .compatible = "fsl,imx8m-ddr-pmu", .data = &imx8m_devtype_data},
 	{ /* sentinel */ }
 };
 
@@ -57,6 +70,7 @@ struct ddr_pmu {
 	struct perf_event *events[NUM_COUNTERS];
 	int active_events;
 	enum cpuhp_state cpuhp_state;
+	const struct fsl_ddr_devtype_data *devtype_data;
 	int irq;
 	int id;
 };
@@ -128,6 +142,8 @@ static struct attribute *ddr_perf_events_attrs[] = {
 	IMX8_DDR_PMU_EVENT_ATTR(refresh, 0x37),
 	IMX8_DDR_PMU_EVENT_ATTR(write, 0x38),
 	IMX8_DDR_PMU_EVENT_ATTR(raw-hazard, 0x39),
+	IMX8_DDR_PMU_EVENT_ATTR(axi-id-read, 0x41),
+	IMX8_DDR_PMU_EVENT_ATTR(axi-id-write, 0x42),
 	NULL,
 };
 
@@ -137,9 +153,11 @@ static struct attribute_group ddr_perf_events_attr_group = {
 };
 
 PMU_FORMAT_ATTR(event, "config:0-7");
+PMU_FORMAT_ATTR(axi_id, "config1:0-31");
 
 static struct attribute *ddr_perf_format_attrs[] = {
 	&format_attr_event.attr,
+	&format_attr_axi_id.attr,
 	NULL,
 };
 
@@ -189,6 +207,16 @@ static u32 ddr_perf_read_counter(struct ddr_pmu *pmu, int counter)
 	return readl_relaxed(pmu->base + COUNTER_READ + counter * 4);
 }
 
+static bool ddr_perf_is_filtered(struct perf_event *event)
+{
+	return event->attr.config == 0x41 || event->attr.config == 0x42;
+}
+
+static u32 ddr_perf_filter_val(struct perf_event *event)
+{
+	return event->attr.config1;
+}
+
 static int ddr_perf_event_init(struct perf_event *event)
 {
 	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
@@ -215,6 +243,18 @@ static int ddr_perf_event_init(struct perf_event *event)
 			!is_software_event(event->group_leader))
 		return -EINVAL;
 
+	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
+		bool is_filtered = ddr_perf_is_filtered(event);
+		u32 filter_val = ddr_perf_filter_val(event);
+
+		for_each_sibling_event(sibling, event->group_leader) {
+			if (is_filtered && ddr_perf_is_filtered(sibling) &&
+			    ddr_perf_filter_val(sibling) != filter_val) {
+				return -EINVAL;
+			}
+		}
+	}
+
 	for_each_sibling_event(sibling, event->group_leader) {
 		if (sibling->pmu != event->pmu &&
 				!is_software_event(sibling))
@@ -288,6 +328,23 @@ static int ddr_perf_event_add(struct perf_event *event, int flags)
 	int counter;
 	int cfg = event->attr.config;
 
+	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
+		int i;
+		bool is_filtered = ddr_perf_is_filtered(event);
+		u32 filter_val = ddr_perf_filter_val(event);
+
+		for (i = 1; i < NUM_COUNTERS; i++) {
+			if (is_filtered && pmu->events[i] &&
+			    ddr_perf_is_filtered(pmu->events[i]) &&
+			    ddr_perf_filter_val(pmu->events[i]) != filter_val) {
+				dev_dbg(pmu->dev, "Contradictory axi id filter value\n");
+				return -EINVAL;
+			}
+		}
+
+		writel(filter_val, pmu->base + COUNTER_DPCR1);
+	}
+
 	counter = ddr_perf_alloc_counter(pmu, cfg);
 	if (counter < 0) {
 		dev_dbg(pmu->dev, "There are not enough counters\n");
@@ -472,6 +529,8 @@ static int ddr_perf_probe(struct platform_device *pdev)
 	if (!name)
 		return -ENOMEM;
 
+	pmu->devtype_data = of_device_get_match_data(&pdev->dev);
+
 	pmu->cpu = raw_smp_processor_id();
 	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
 				      DDR_CPUHP_CB_NAME,
-- 
2.17.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

[PATCH V5 2/3] Documentation: admin-guide: perf: add i.MX8 ddr pmu user doc

[Joakim Zhang]

Add i.MX8 ddr pmu user doc.

ChangeLog:
V1 -> V4:
	* new add in V4.
V4 -> V5:
	* no change.

Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
---
 Documentation/admin-guide/perf/imx-ddr.rst | 30 ++++++++++++++++++++++
 1 file changed, 30 insertions(+)
 create mode 100644 Documentation/admin-guide/perf/imx-ddr.rst

diff --git a/Documentation/admin-guide/perf/imx-ddr.rst b/Documentation/admin-guide/perf/imx-ddr.rst
new file mode 100644
index 000000000000..8ab50e2da27e
--- /dev/null
+++ b/Documentation/admin-guide/perf/imx-ddr.rst
@@ -0,0 +1,30 @@
+====================================================
+Freescale i.MX8 DDR Performance Monitoring Unit (PMU)
+====================================================
+
+There are no performance counters inside the DRAM controller, so performance
+signals are brought out to the edge of the controller where a set of 4 x 32 bit
+counters is implemented. This is controlled by the Performance log on parameter
+which causes a large number of PERF signals to be generated.
+
+Selection of the value for each counter is done via the config registiers. There
+is one register for each counter. Counter 0 is special in that it always counts
+“time” and when expired causes a lock on itself and the other counters and an
+interrupt ie enable of counter 0 is a global function.
+
+The "format" directory describes format of the config (event ID) and config1
+(AXI ID filter) fields of the perf_event_attr structure, see /sys/bus/event_source/
+devices/imx8_ddr0/format/. The "events" directory describes the events types
+hardware supported that can be used with perf tool, see /sys/bus/event_source/
+devices/imx8_ddr0/events/.
+
+AXI ID filter is only used by CSV modes 0x41 (axi-id-read) and 0x42 (axi-id-write)
+to count reading or writing matches filter setting. User should specify this two
+events with the same AXI ID filter value if want to count at the same time, as
+this filter register is shared between counters.
+
+Example for perf tool use::
+
+        perf stat -e imx8_ddr0/cycles/ sleep 1
+        perf stat -e imx8_ddr0/read/,imx8_ddr0/write/ sleep 1
+        perf stat -e imx8_ddr0/axi-id-read,axi_id=0xMMMMDDDD/,imx8_ddr0/axi-id-write,axi_id=0xMMMMDDDD/ sleep 1
-- 
2.17.1

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

[PATCH V5 3/3] MAINTAINERS: add imx8 ddr perf admin-guide maintainer information

[Joakim Zhang]

Add imx8 ddr perf admin-guide maintainer information.

ChangeLog:
V1 -> V5:
	* new add in V5.

Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
---
 MAINTAINERS | 1 +
 1 file changed, 1 insertion(+)

diff --git a/MAINTAINERS b/MAINTAINERS
index c6fa7e88a6f0..bc4eae7eba64 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -6383,6 +6383,7 @@ M:	Frank Li <Frank.li@nxp.com>
 L:	linux-arm-kernel@lists.infradead.org
 S:	Maintained
 F:	drivers/perf/fsl_imx8_ddr_perf.c
+F:	Documentation/admin-guide/perf/imx-ddr.rst
 F:	Documentation/devicetree/bindings/perf/fsl-imx-ddr.txt
 
 FREESCALE IMX LPI2C DRIVER
-- 
2.17.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

RE: [PATCH V5 0/3] perf: imx8_ddr_perf: add AXI ID filter

[Frank Li]

> -----Original Message-----
> From: Joakim Zhang
> Sent: Thursday, August 8, 2019 1:45 AM
> To: robin.murphy@arm.com; will@kernel.org; mark.rutland@arm.com
> Cc: Frank Li <frank.li@nxp.com>; linux-arm-kernel@lists.infradead.org; dl-linux-
> imx <linux-imx@nxp.com>; Joakim Zhang <qiangqing.zhang@nxp.com>
> Subject: [PATCH V5 0/3] perf: imx8_ddr_perf: add AXI ID filter
> 
> Add AXI ID filter for imx8m ddr perf.
> 
> Joakim Zhang (3):
>   perf: imx8_ddr_perf: add AXI ID filter support
>   Documentation: admin-guide: perf: add i.MX8 ddr pmu user doc
>   MAINTAINERS: add imx8 ddr perf admin-guide maintainer information
> 
>  Documentation/admin-guide/perf/imx-ddr.rst | 30 +++++++++++
>  MAINTAINERS                                |  1 +
>  drivers/perf/fsl_imx8_ddr_perf.c           | 63 +++++++++++++++++++++-
>  3 files changed, 92 insertions(+), 2 deletions(-)  create mode 100644
> Documentation/admin-guide/perf/imx-ddr.rst
> 
Acked-by: Frank Li <Frank.li@nxp.com>

> --
> 2.17.1


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

RE: [PATCH V5 0/3] perf: imx8_ddr_perf: add AXI ID filter

[Joakim Zhang]

Kindly ping...

Best Regards,
Joakim Zhang

> -----Original Message-----
> From: Frank Li
> Sent: 2019年8月8日 21:56
> To: Joakim Zhang <qiangqing.zhang@nxp.com>; robin.murphy@arm.com;
> will@kernel.org; mark.rutland@arm.com
> Cc: linux-arm-kernel@lists.infradead.org; dl-linux-imx <linux-imx@nxp.com>
> Subject: RE: [PATCH V5 0/3] perf: imx8_ddr_perf: add AXI ID filter
> 
> 
> 
> > -----Original Message-----
> > From: Joakim Zhang
> > Sent: Thursday, August 8, 2019 1:45 AM
> > To: robin.murphy@arm.com; will@kernel.org; mark.rutland@arm.com
> > Cc: Frank Li <frank.li@nxp.com>; linux-arm-kernel@lists.infradead.org;
> > dl-linux- imx <linux-imx@nxp.com>; Joakim Zhang
> > <qiangqing.zhang@nxp.com>
> > Subject: [PATCH V5 0/3] perf: imx8_ddr_perf: add AXI ID filter
> >
> > Add AXI ID filter for imx8m ddr perf.
> >
> > Joakim Zhang (3):
> >   perf: imx8_ddr_perf: add AXI ID filter support
> >   Documentation: admin-guide: perf: add i.MX8 ddr pmu user doc
> >   MAINTAINERS: add imx8 ddr perf admin-guide maintainer information
> >
> >  Documentation/admin-guide/perf/imx-ddr.rst | 30 +++++++++++
> >  MAINTAINERS                                |  1 +
> >  drivers/perf/fsl_imx8_ddr_perf.c           | 63
> +++++++++++++++++++++-
> >  3 files changed, 92 insertions(+), 2 deletions(-)  create mode 100644
> > Documentation/admin-guide/perf/imx-ddr.rst
> >
> Acked-by: Frank Li <Frank.li@nxp.com>
> 
> > --
> > 2.17.1

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter support

[Mark Rutland]

On Thu, Aug 08, 2019 at 06:45:29AM +0000, Joakim Zhang wrote:
> AXI filtering is used by CSV modes 0x41 and 0x42 to count reads or
> writes with an ARID or AXID matching filter setting. Granularity is at
> subsystem level. Implementation does not allow filtring between masters
> within a subsystem. Filter is defined with 2 configuration registers.
> 
> --AXI_ID defines AxID matching value
> --AXI_MASKING defines which bits of AxID are meaningful for the matching
> 
> When non-masked bits are matching corresponding AXI_ID bits then counter
> is incremented. This filter allows counting read or write access from a
> subsystem or multiple subsystems.
> 
> Perf counter is incremented if AxID && AXI_MASKING == AXI_ID && AXI_MASKING

Just to check, the filter does not affect other events, right?

> 
> AXI_ID and AXI_MASKING are mapped on DPCR1 register in performance counter.
> 
> Read and write AXI ID filter should write same value to DPCR1 if want to
> specify at the same time as this filter is shared between counters.
> 
> e.g.
> perf stat -a -e imx8_ddr0/axi-id-read,axi_id=0xMMMMDDDD/,imx8_ddr0/axi-id-write,axi_id=0xMMMMDDDD/ cmd
> MMMM: AXI_MASKING
> DDDD: AXI_ID

You might want to expose this to userspace as two fields:

	axi_id=DDDD
	axi_mask=MMMM

... where axi_mask is inverted (i.e. set bits are bits to ignore), so
that the user can just specify axi_id to monitor a specific id, rather
than having to specifiy axi_id=0xffff<id>.

> 
> ChangeLog:
> V1 -> V2:
> 	* add error log if user specifies read/write AXI ID filter at
> 	the same time.
> 	* of_device_get_match_data() instead of of_match_device(), and
> 	remove the check of return value.
> V2 -> V3:
> 	* move the AXI ID check to event_add().
> 	* add support for same value of axi_id.
> V3 -> V4:
> 	* move the AXI ID check to event_init().
> V4 -> V5:
> 	* reject event group if AXI ID not consistent in event_init().
> 
> Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
> ---
>  drivers/perf/fsl_imx8_ddr_perf.c | 63 +++++++++++++++++++++++++++++++-
>  1 file changed, 61 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/perf/fsl_imx8_ddr_perf.c b/drivers/perf/fsl_imx8_ddr_perf.c
> index 63fe21600072..f25cf5cbe156 100644
> --- a/drivers/perf/fsl_imx8_ddr_perf.c
> +++ b/drivers/perf/fsl_imx8_ddr_perf.c
> @@ -42,9 +42,22 @@
>  
>  static DEFINE_IDA(ddr_ida);
>  
> +/* DDR Perf hardware feature */
> +#define DDR_CAP_AXI_ID_FILTER		0x1	/* support AXI ID filter */
> +
> +struct fsl_ddr_devtype_data {
> +	unsigned int quirks;	/* quirks needed for different DDR Perf core */
> +};
> +
> +static const struct fsl_ddr_devtype_data imx8_devtype_data;
> +
> +static const struct fsl_ddr_devtype_data imx8m_devtype_data = {
> +	.quirks = DDR_CAP_AXI_ID_FILTER,
> +};
> +
>  static const struct of_device_id imx_ddr_pmu_dt_ids[] = {
> -	{ .compatible = "fsl,imx8-ddr-pmu",},
> -	{ .compatible = "fsl,imx8m-ddr-pmu",},
> +	{ .compatible = "fsl,imx8-ddr-pmu", .data = &imx8_devtype_data},
> +	{ .compatible = "fsl,imx8m-ddr-pmu", .data = &imx8m_devtype_data},
>  	{ /* sentinel */ }
>  };

Are new DDR PMUs going to lack this feature?

If all PMUs the driver supports have it, then we don't need this
quirk/feature list.

That would make the subsequent code a bit nicer, as all the filtering
code would lose a level of indentation.

>  
> @@ -57,6 +70,7 @@ struct ddr_pmu {
>  	struct perf_event *events[NUM_COUNTERS];
>  	int active_events;
>  	enum cpuhp_state cpuhp_state;
> +	const struct fsl_ddr_devtype_data *devtype_data;
>  	int irq;
>  	int id;
>  };
> @@ -128,6 +142,8 @@ static struct attribute *ddr_perf_events_attrs[] = {
>  	IMX8_DDR_PMU_EVENT_ATTR(refresh, 0x37),
>  	IMX8_DDR_PMU_EVENT_ATTR(write, 0x38),
>  	IMX8_DDR_PMU_EVENT_ATTR(raw-hazard, 0x39),
> +	IMX8_DDR_PMU_EVENT_ATTR(axi-id-read, 0x41),
> +	IMX8_DDR_PMU_EVENT_ATTR(axi-id-write, 0x42),
>  	NULL,
>  };
>  
> @@ -137,9 +153,11 @@ static struct attribute_group ddr_perf_events_attr_group = {
>  };
>  
>  PMU_FORMAT_ATTR(event, "config:0-7");
> +PMU_FORMAT_ATTR(axi_id, "config1:0-31");
>  
>  static struct attribute *ddr_perf_format_attrs[] = {
>  	&format_attr_event.attr,
> +	&format_attr_axi_id.attr,
>  	NULL,
>  };
>  
> @@ -189,6 +207,16 @@ static u32 ddr_perf_read_counter(struct ddr_pmu *pmu, int counter)
>  	return readl_relaxed(pmu->base + COUNTER_READ + counter * 4);
>  }
>  
> +static bool ddr_perf_is_filtered(struct perf_event *event)
> +{
> +	return event->attr.config == 0x41 || event->attr.config == 0x42;
> +}
> +
> +static u32 ddr_perf_filter_val(struct perf_event *event)
> +{
> +	return event->attr.config1;
> +}
> +

Let's add another helper:

static bool ddr_perf_filters_compatible(struct perf_event *a,
					struct perf_event *b)
{
	if (!ddr_perf_is_filtered(a))
		return true;
	if (!ddr_perf_is_filtered(b))
		return true;
	return ddr_perf_filter_val(a) == ddr_perf_filter_val(b);
}

>  static int ddr_perf_event_init(struct perf_event *event)
>  {
>  	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
> @@ -215,6 +243,18 @@ static int ddr_perf_event_init(struct perf_event *event)
>  			!is_software_event(event->group_leader))
>  		return -EINVAL;
>  
> +	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> +		bool is_filtered = ddr_perf_is_filtered(event);
> +		u32 filter_val = ddr_perf_filter_val(event);
> +
> +		for_each_sibling_event(sibling, event->group_leader) {
> +			if (is_filtered && ddr_perf_is_filtered(sibling) &&
> +			    ddr_perf_filter_val(sibling) != filter_val) {
> +				return -EINVAL;
> +			}
> +		}
> +	}

... so this can be:

	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
		if (!ddr_perf_filters_compatible(event, event->group_leader))
			return -EINVAL;
		for_each_sibling_event(sibling, event->group_leader) {
			if (!ddr_perf_filters_compatible(event, sibling))
				return -EINVAL;
		}
	}

Note: that checks group_leader, which you mised above. When the event is
the group leader, it's trivially compatible with itself, so we don't
need a special case there.

> +
>  	for_each_sibling_event(sibling, event->group_leader) {
>  		if (sibling->pmu != event->pmu &&
>  				!is_software_event(sibling))
> @@ -288,6 +328,23 @@ static int ddr_perf_event_add(struct perf_event *event, int flags)
>  	int counter;
>  	int cfg = event->attr.config;
>  
> +	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> +		int i;
> +		bool is_filtered = ddr_perf_is_filtered(event);
> +		u32 filter_val = ddr_perf_filter_val(event);
> +
> +		for (i = 1; i < NUM_COUNTERS; i++) {
> +			if (is_filtered && pmu->events[i] &&
> +			    ddr_perf_is_filtered(pmu->events[i]) &&
> +			    ddr_perf_filter_val(pmu->events[i]) != filter_val) {
> +				dev_dbg(pmu->dev, "Contradictory axi id filter value\n");
> +				return -EINVAL;
> +			}
> +		}

... and likewise:

	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
		int i;

		for (i = 1; i < NUM_COUNTERS; i++) {
			if (!ddr_perf_filters_compatible(event, pmu->events[i]))
				return -EINVAL;
		}
	}

Please drop the dev_dbg() here, since when perf rotates events this can
happen many times per second, and it's entirely legitimate.

Otherwise, this looks good to me.

Thanks,
Mark.

> +
> +		writel(filter_val, pmu->base + COUNTER_DPCR1);
> +	}
> +
>  	counter = ddr_perf_alloc_counter(pmu, cfg);
>  	if (counter < 0) {
>  		dev_dbg(pmu->dev, "There are not enough counters\n");
> @@ -472,6 +529,8 @@ static int ddr_perf_probe(struct platform_device *pdev)
>  	if (!name)
>  		return -ENOMEM;
>  
> +	pmu->devtype_data = of_device_get_match_data(&pdev->dev);
> +
>  	pmu->cpu = raw_smp_processor_id();
>  	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
>  				      DDR_CPUHP_CB_NAME,
> -- 
> 2.17.1
> 

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

RE: [PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter support

[Joakim Zhang]

> -----Original Message-----
> From: Mark Rutland <mark.rutland@arm.com>
> Sent: 2019年8月23日 20:57
> To: Joakim Zhang <qiangqing.zhang@nxp.com>
> Cc: robin.murphy@arm.com; will@kernel.org; Frank Li <frank.li@nxp.com>;
> linux-arm-kernel@lists.infradead.org; dl-linux-imx <linux-imx@nxp.com>
> Subject: Re: [PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter support
> 
> On Thu, Aug 08, 2019 at 06:45:29AM +0000, Joakim Zhang wrote:
> > AXI filtering is used by CSV modes 0x41 and 0x42 to count reads or
> > writes with an ARID or AXID matching filter setting. Granularity is at
> > subsystem level. Implementation does not allow filtring between
> > masters within a subsystem. Filter is defined with 2 configuration registers.
> >
> > --AXI_ID defines AxID matching value
> > --AXI_MASKING defines which bits of AxID are meaningful for the
> > matching
> >
> > When non-masked bits are matching corresponding AXI_ID bits then
> > counter is incremented. This filter allows counting read or write
> > access from a subsystem or multiple subsystems.
> >
> > Perf counter is incremented if AxID && AXI_MASKING == AXI_ID &&
> > AXI_MASKING
> 
> Just to check, the filter does not affect other events, right?

[Joakim] Yes, this filter is only for events 0x41 and 0x42.

> >
> > AXI_ID and AXI_MASKING are mapped on DPCR1 register in performance
> counter.
> >
> > Read and write AXI ID filter should write same value to DPCR1 if want
> > to specify at the same time as this filter is shared between counters.
> >
> > e.g.
> > perf stat -a -e
> >
> imx8_ddr0/axi-id-read,axi_id=0xMMMMDDDD/,imx8_ddr0/axi-id-write,axi_id
> > =0xMMMMDDDD/ cmd
> > MMMM: AXI_MASKING
> > DDDD: AXI_ID
> 
> You might want to expose this to userspace as two fields:
> 
> 	axi_id=DDDD
> 	axi_mask=MMMM
> 
> ... where axi_mask is inverted (i.e. set bits are bits to ignore), so that the user
> can just specify axi_id to monitor a specific id, rather than having to specifiy
> axi_id=0xffff<id>.

[Joakim] No, axi_mask is not inverted, should specify axi_id = 0xffff<id> for the particular AXI ID. I will improve the commit message.
		0: corresponding bit is masked
		1: corresponding bit is not masked, i.e. used to do the matching

> >
> > ChangeLog:
> > V1 -> V2:
> > 	* add error log if user specifies read/write AXI ID filter at
> > 	the same time.
> > 	* of_device_get_match_data() instead of of_match_device(), and
> > 	remove the check of return value.
> > V2 -> V3:
> > 	* move the AXI ID check to event_add().
> > 	* add support for same value of axi_id.
> > V3 -> V4:
> > 	* move the AXI ID check to event_init().
> > V4 -> V5:
> > 	* reject event group if AXI ID not consistent in event_init().
> >
> > Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
> > ---
> >  drivers/perf/fsl_imx8_ddr_perf.c | 63
> > +++++++++++++++++++++++++++++++-
> >  1 file changed, 61 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/perf/fsl_imx8_ddr_perf.c
> > b/drivers/perf/fsl_imx8_ddr_perf.c
> > index 63fe21600072..f25cf5cbe156 100644
> > --- a/drivers/perf/fsl_imx8_ddr_perf.c
> > +++ b/drivers/perf/fsl_imx8_ddr_perf.c
> > @@ -42,9 +42,22 @@
> >
> >  static DEFINE_IDA(ddr_ida);
> >
> > +/* DDR Perf hardware feature */
> > +#define DDR_CAP_AXI_ID_FILTER		0x1	/* support AXI ID filter */
> > +
> > +struct fsl_ddr_devtype_data {
> > +	unsigned int quirks;	/* quirks needed for different DDR Perf core */
> > +};
> > +
> > +static const struct fsl_ddr_devtype_data imx8_devtype_data;
> > +
> > +static const struct fsl_ddr_devtype_data imx8m_devtype_data = {
> > +	.quirks = DDR_CAP_AXI_ID_FILTER,
> > +};
> > +
> >  static const struct of_device_id imx_ddr_pmu_dt_ids[] = {
> > -	{ .compatible = "fsl,imx8-ddr-pmu",},
> > -	{ .compatible = "fsl,imx8m-ddr-pmu",},
> > +	{ .compatible = "fsl,imx8-ddr-pmu", .data = &imx8_devtype_data},
> > +	{ .compatible = "fsl,imx8m-ddr-pmu", .data = &imx8m_devtype_data},
> >  	{ /* sentinel */ }
> >  };
> 
> Are new DDR PMUs going to lack this feature?
> 
> If all PMUs the driver supports have it, then we don't need this quirk/feature
> list.
> 
> That would make the subsequent code a bit nicer, as all the filtering code would
> lose a level of indentation.

[Joakim] This feature may drop, some coming SOCs will improve this AXI ID filtering, let it can filter from different ID separately, and ang other extensions.
 
> >
> > @@ -57,6 +70,7 @@ struct ddr_pmu {
> >  	struct perf_event *events[NUM_COUNTERS];
> >  	int active_events;
> >  	enum cpuhp_state cpuhp_state;
> > +	const struct fsl_ddr_devtype_data *devtype_data;
> >  	int irq;
> >  	int id;
> >  };
> > @@ -128,6 +142,8 @@ static struct attribute *ddr_perf_events_attrs[] = {
> >  	IMX8_DDR_PMU_EVENT_ATTR(refresh, 0x37),
> >  	IMX8_DDR_PMU_EVENT_ATTR(write, 0x38),
> >  	IMX8_DDR_PMU_EVENT_ATTR(raw-hazard, 0x39),
> > +	IMX8_DDR_PMU_EVENT_ATTR(axi-id-read, 0x41),
> > +	IMX8_DDR_PMU_EVENT_ATTR(axi-id-write, 0x42),
> >  	NULL,
> >  };
> >
> > @@ -137,9 +153,11 @@ static struct attribute_group
> > ddr_perf_events_attr_group = {  };
> >
> >  PMU_FORMAT_ATTR(event, "config:0-7");
> > +PMU_FORMAT_ATTR(axi_id, "config1:0-31");
> >
> >  static struct attribute *ddr_perf_format_attrs[] = {
> >  	&format_attr_event.attr,
> > +	&format_attr_axi_id.attr,
> >  	NULL,
> >  };
> >
> > @@ -189,6 +207,16 @@ static u32 ddr_perf_read_counter(struct ddr_pmu
> *pmu, int counter)
> >  	return readl_relaxed(pmu->base + COUNTER_READ + counter * 4);  }
> >
> > +static bool ddr_perf_is_filtered(struct perf_event *event) {
> > +	return event->attr.config == 0x41 || event->attr.config == 0x42; }
> > +
> > +static u32 ddr_perf_filter_val(struct perf_event *event) {
> > +	return event->attr.config1;
> > +}
> > +
> 
> Let's add another helper:
> 
> static bool ddr_perf_filters_compatible(struct perf_event *a,
> 					struct perf_event *b)
> {
> 	if (!ddr_perf_is_filtered(a))
> 		return true;
> 	if (!ddr_perf_is_filtered(b))
> 		return true;
> 	return ddr_perf_filter_val(a) == ddr_perf_filter_val(b); }
> 
> >  static int ddr_perf_event_init(struct perf_event *event)  {
> >  	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu); @@ -215,6 +243,18
> @@
> > static int ddr_perf_event_init(struct perf_event *event)
> >  			!is_software_event(event->group_leader))
> >  		return -EINVAL;
> >
> > +	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> > +		bool is_filtered = ddr_perf_is_filtered(event);
> > +		u32 filter_val = ddr_perf_filter_val(event);
> > +
> > +		for_each_sibling_event(sibling, event->group_leader) {
> > +			if (is_filtered && ddr_perf_is_filtered(sibling) &&
> > +			    ddr_perf_filter_val(sibling) != filter_val) {
> > +				return -EINVAL;
> > +			}
> > +		}
> > +	}
> 
> ... so this can be:
> 
> 	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> 		if (!ddr_perf_filters_compatible(event, event->group_leader))
> 			return -EINVAL;
> 		for_each_sibling_event(sibling, event->group_leader) {
> 			if (!ddr_perf_filters_compatible(event, sibling))
> 				return -EINVAL;
> 		}
> 	}

[Joakim] Got it.

> Note: that checks group_leader, which you mised above. When the event is the
> group leader, it's trivially compatible with itself, so we don't need a special case
> there.
> 
> > +
> >  	for_each_sibling_event(sibling, event->group_leader) {
> >  		if (sibling->pmu != event->pmu &&
> >  				!is_software_event(sibling))
> > @@ -288,6 +328,23 @@ static int ddr_perf_event_add(struct perf_event
> *event, int flags)
> >  	int counter;
> >  	int cfg = event->attr.config;
> >
> > +	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> > +		int i;
> > +		bool is_filtered = ddr_perf_is_filtered(event);
> > +		u32 filter_val = ddr_perf_filter_val(event);
> > +
> > +		for (i = 1; i < NUM_COUNTERS; i++) {
> > +			if (is_filtered && pmu->events[i] &&
> > +			    ddr_perf_is_filtered(pmu->events[i]) &&
> > +			    ddr_perf_filter_val(pmu->events[i]) != filter_val) {
> > +				dev_dbg(pmu->dev, "Contradictory axi id filter value\n");
> > +				return -EINVAL;
> > +			}
> > +		}
> 
> ... and likewise:
> 
> 	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> 		int i;
> 
> 		for (i = 1; i < NUM_COUNTERS; i++) {
> 			if (!ddr_perf_filters_compatible(event, pmu->events[i]))
> 				return -EINVAL;
> 		}
> 	}
> 
> Please drop the dev_dbg() here, since when perf rotates events this can
> happen many times per second, and it's entirely legitimate.

[Joakim] Got it.

> Otherwise, this looks good to me.

[Joakim] Thanks Mark, I will sent out a V6, please help review.

Best Regards,
Joakim Zhang
> Thanks,
> Mark.
> 
> > +
> > +		writel(filter_val, pmu->base + COUNTER_DPCR1);
> > +	}
> > +
> >  	counter = ddr_perf_alloc_counter(pmu, cfg);
> >  	if (counter < 0) {
> >  		dev_dbg(pmu->dev, "There are not enough counters\n"); @@ -472,6
> > +529,8 @@ static int ddr_perf_probe(struct platform_device *pdev)
> >  	if (!name)
> >  		return -ENOMEM;
> >
> > +	pmu->devtype_data = of_device_get_match_data(&pdev->dev);
> > +
> >  	pmu->cpu = raw_smp_processor_id();
> >  	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
> >  				      DDR_CPUHP_CB_NAME,
> > --
> > 2.17.1
> >
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

RE: [PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter support

[Frank Li]

> -----Original Message-----
> From: Joakim Zhang
> Sent: Monday, August 26, 2019 2:12 AM
> To: Mark Rutland <mark.rutland@arm.com>
> Cc: robin.murphy@arm.com; will@kernel.org; Frank Li <frank.li@nxp.com>;
> linux-arm-kernel@lists.infradead.org; dl-linux-imx <linux-imx@nxp.com>
> Subject: RE: [PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter support
> 
> 
> > -----Original Message-----
> > From: Mark Rutland <mark.rutland@arm.com>
> > Sent: 2019年8月23日 20:57
> > To: Joakim Zhang <qiangqing.zhang@nxp.com>
> > Cc: robin.murphy@arm.com; will@kernel.org; Frank Li
> > <frank.li@nxp.com>; linux-arm-kernel@lists.infradead.org; dl-linux-imx
> > <linux-imx@nxp.com>
> > Subject: Re: [PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter
> > support
> >
> > On Thu, Aug 08, 2019 at 06:45:29AM +0000, Joakim Zhang wrote:
> > > AXI filtering is used by CSV modes 0x41 and 0x42 to count reads or
> > > writes with an ARID or AXID matching filter setting. Granularity is
> > > at subsystem level. Implementation does not allow filtring between
> > > masters within a subsystem. Filter is defined with 2 configuration registers.
> > >
> > > --AXI_ID defines AxID matching value --AXI_MASKING defines which
> > > bits of AxID are meaningful for the matching
> > >
> > > When non-masked bits are matching corresponding AXI_ID bits then
> > > counter is incremented. This filter allows counting read or write
> > > access from a subsystem or multiple subsystems.
> > >
> > > Perf counter is incremented if AxID && AXI_MASKING == AXI_ID &&
> > > AXI_MASKING
> >
> > Just to check, the filter does not affect other events, right?
> 
> [Joakim] Yes, this filter is only for events 0x41 and 0x42.
> 
> > >
> > > AXI_ID and AXI_MASKING are mapped on DPCR1 register in performance
> > counter.
> > >
> > > Read and write AXI ID filter should write same value to DPCR1 if
> > > want to specify at the same time as this filter is shared between counters.
> > >
> > > e.g.
> > > perf stat -a -e
> > >
> > imx8_ddr0/axi-id-read,axi_id=0xMMMMDDDD/,imx8_ddr0/axi-id-write,axi_id
> > > =0xMMMMDDDD/ cmd
> > > MMMM: AXI_MASKING
> > > DDDD: AXI_ID
> >
> > You might want to expose this to userspace as two fields:
> >
> > 	axi_id=DDDD
> > 	axi_mask=MMMM
> >
> > ... where axi_mask is inverted (i.e. set bits are bits to ignore), so
> > that the user can just specify axi_id to monitor a specific id, rather
> > than having to specifiy axi_id=0xffff<id>.
> 
> [Joakim] No, axi_mask is not inverted, should specify axi_id = 0xffff<id> for the
> particular AXI ID. I will improve the commit message.
> 		0: corresponding bit is masked
> 		1: corresponding bit is not masked, i.e. used to do the matching
> 

 [Frank Li] Joakim,  mark's means is that revert it at your driver.
Most case user just want to filter a AXID. For example 0x12

If you revert mask at driver,  user just need input axi_id = 0x12 instead of axi_id=0xffff0012.


> > >
> > > ChangeLog:
> > > V1 -> V2:
> > > 	* add error log if user specifies read/write AXI ID filter at
> > > 	the same time.
> > > 	* of_device_get_match_data() instead of of_match_device(), and
> > > 	remove the check of return value.
> > > V2 -> V3:
> > > 	* move the AXI ID check to event_add().
> > > 	* add support for same value of axi_id.
> > > V3 -> V4:
> > > 	* move the AXI ID check to event_init().
> > > V4 -> V5:
> > > 	* reject event group if AXI ID not consistent in event_init().
> > >
> > > Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
> > > ---
> > >  drivers/perf/fsl_imx8_ddr_perf.c | 63
> > > +++++++++++++++++++++++++++++++-
> > >  1 file changed, 61 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/perf/fsl_imx8_ddr_perf.c
> > > b/drivers/perf/fsl_imx8_ddr_perf.c
> > > index 63fe21600072..f25cf5cbe156 100644
> > > --- a/drivers/perf/fsl_imx8_ddr_perf.c
> > > +++ b/drivers/perf/fsl_imx8_ddr_perf.c
> > > @@ -42,9 +42,22 @@
> > >
> > >  static DEFINE_IDA(ddr_ida);
> > >
> > > +/* DDR Perf hardware feature */
> > > +#define DDR_CAP_AXI_ID_FILTER		0x1	/* support AXI ID filter
> */
> > > +
> > > +struct fsl_ddr_devtype_data {
> > > +	unsigned int quirks;	/* quirks needed for different DDR Perf core */
> > > +};
> > > +
> > > +static const struct fsl_ddr_devtype_data imx8_devtype_data;
> > > +
> > > +static const struct fsl_ddr_devtype_data imx8m_devtype_data = {
> > > +	.quirks = DDR_CAP_AXI_ID_FILTER,
> > > +};
> > > +
> > >  static const struct of_device_id imx_ddr_pmu_dt_ids[] = {
> > > -	{ .compatible = "fsl,imx8-ddr-pmu",},
> > > -	{ .compatible = "fsl,imx8m-ddr-pmu",},
> > > +	{ .compatible = "fsl,imx8-ddr-pmu", .data = &imx8_devtype_data},
> > > +	{ .compatible = "fsl,imx8m-ddr-pmu", .data = &imx8m_devtype_data},
> > >  	{ /* sentinel */ }
> > >  };
> >
> > Are new DDR PMUs going to lack this feature?
> >
> > If all PMUs the driver supports have it, then we don't need this
> > quirk/feature list.
> >
> > That would make the subsequent code a bit nicer, as all the filtering
> > code would lose a level of indentation.
> 
> [Joakim] This feature may drop, some coming SOCs will improve this AXI ID
> filtering, let it can filter from different ID separately, and ang other extensions.
> 
> > >
> > > @@ -57,6 +70,7 @@ struct ddr_pmu {
> > >  	struct perf_event *events[NUM_COUNTERS];
> > >  	int active_events;
> > >  	enum cpuhp_state cpuhp_state;
> > > +	const struct fsl_ddr_devtype_data *devtype_data;
> > >  	int irq;
> > >  	int id;
> > >  };
> > > @@ -128,6 +142,8 @@ static struct attribute *ddr_perf_events_attrs[] = {
> > >  	IMX8_DDR_PMU_EVENT_ATTR(refresh, 0x37),
> > >  	IMX8_DDR_PMU_EVENT_ATTR(write, 0x38),
> > >  	IMX8_DDR_PMU_EVENT_ATTR(raw-hazard, 0x39),
> > > +	IMX8_DDR_PMU_EVENT_ATTR(axi-id-read, 0x41),
> > > +	IMX8_DDR_PMU_EVENT_ATTR(axi-id-write, 0x42),
> > >  	NULL,
> > >  };
> > >
> > > @@ -137,9 +153,11 @@ static struct attribute_group
> > > ddr_perf_events_attr_group = {  };
> > >
> > >  PMU_FORMAT_ATTR(event, "config:0-7");
> > > +PMU_FORMAT_ATTR(axi_id, "config1:0-31");
> > >
> > >  static struct attribute *ddr_perf_format_attrs[] = {
> > >  	&format_attr_event.attr,
> > > +	&format_attr_axi_id.attr,
> > >  	NULL,
> > >  };
> > >
> > > @@ -189,6 +207,16 @@ static u32 ddr_perf_read_counter(struct ddr_pmu
> > *pmu, int counter)
> > >  	return readl_relaxed(pmu->base + COUNTER_READ + counter * 4);  }
> > >
> > > +static bool ddr_perf_is_filtered(struct perf_event *event) {
> > > +	return event->attr.config == 0x41 || event->attr.config == 0x42; }
> > > +
> > > +static u32 ddr_perf_filter_val(struct perf_event *event) {
> > > +	return event->attr.config1;
> > > +}
> > > +
> >
> > Let's add another helper:
> >
> > static bool ddr_perf_filters_compatible(struct perf_event *a,
> > 					struct perf_event *b)
> > {
> > 	if (!ddr_perf_is_filtered(a))
> > 		return true;
> > 	if (!ddr_perf_is_filtered(b))
> > 		return true;
> > 	return ddr_perf_filter_val(a) == ddr_perf_filter_val(b); }
> >
> > >  static int ddr_perf_event_init(struct perf_event *event)  {
> > >  	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu); @@ -215,6 +243,18
> > @@
> > > static int ddr_perf_event_init(struct perf_event *event)
> > >  			!is_software_event(event->group_leader))
> > >  		return -EINVAL;
> > >
> > > +	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> > > +		bool is_filtered = ddr_perf_is_filtered(event);
> > > +		u32 filter_val = ddr_perf_filter_val(event);
> > > +
> > > +		for_each_sibling_event(sibling, event->group_leader) {
> > > +			if (is_filtered && ddr_perf_is_filtered(sibling) &&
> > > +			    ddr_perf_filter_val(sibling) != filter_val) {
> > > +				return -EINVAL;
> > > +			}
> > > +		}
> > > +	}
> >
> > ... so this can be:
> >
> > 	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> > 		if (!ddr_perf_filters_compatible(event, event->group_leader))
> > 			return -EINVAL;
> > 		for_each_sibling_event(sibling, event->group_leader) {
> > 			if (!ddr_perf_filters_compatible(event, sibling))
> > 				return -EINVAL;
> > 		}
> > 	}
> 
> [Joakim] Got it.
> 
> > Note: that checks group_leader, which you mised above. When the event
> > is the group leader, it's trivially compatible with itself, so we
> > don't need a special case there.
> >
> > > +
> > >  	for_each_sibling_event(sibling, event->group_leader) {
> > >  		if (sibling->pmu != event->pmu &&
> > >  				!is_software_event(sibling))
> > > @@ -288,6 +328,23 @@ static int ddr_perf_event_add(struct perf_event
> > *event, int flags)
> > >  	int counter;
> > >  	int cfg = event->attr.config;
> > >
> > > +	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> > > +		int i;
> > > +		bool is_filtered = ddr_perf_is_filtered(event);
> > > +		u32 filter_val = ddr_perf_filter_val(event);
> > > +
> > > +		for (i = 1; i < NUM_COUNTERS; i++) {
> > > +			if (is_filtered && pmu->events[i] &&
> > > +			    ddr_perf_is_filtered(pmu->events[i]) &&
> > > +			    ddr_perf_filter_val(pmu->events[i]) != filter_val) {
> > > +				dev_dbg(pmu->dev, "Contradictory axi id filter
> value\n");
> > > +				return -EINVAL;
> > > +			}
> > > +		}
> >
> > ... and likewise:
> >
> > 	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> > 		int i;
> >
> > 		for (i = 1; i < NUM_COUNTERS; i++) {
> > 			if (!ddr_perf_filters_compatible(event, pmu->events[i]))
> > 				return -EINVAL;
> > 		}
> > 	}
> >
> > Please drop the dev_dbg() here, since when perf rotates events this
> > can happen many times per second, and it's entirely legitimate.
> 
> [Joakim] Got it.
> 
> > Otherwise, this looks good to me.
> 
> [Joakim] Thanks Mark, I will sent out a V6, please help review.
> 
> Best Regards,
> Joakim Zhang
> > Thanks,
> > Mark.
> >
> > > +
> > > +		writel(filter_val, pmu->base + COUNTER_DPCR1);
> > > +	}
> > > +
> > >  	counter = ddr_perf_alloc_counter(pmu, cfg);
> > >  	if (counter < 0) {
> > >  		dev_dbg(pmu->dev, "There are not enough counters\n"); @@ -
> 472,6
> > > +529,8 @@ static int ddr_perf_probe(struct platform_device *pdev)
> > >  	if (!name)
> > >  		return -ENOMEM;
> > >
> > > +	pmu->devtype_data = of_device_get_match_data(&pdev->dev);
> > > +
> > >  	pmu->cpu = raw_smp_processor_id();
> > >  	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
> > >  				      DDR_CPUHP_CB_NAME,
> > > --
> > > 2.17.1
> > >
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

RE: [PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter support

[Joakim Zhang]

> -----Original Message-----
> From: Frank Li
> Sent: 2019年8月26日 23:37
> To: Joakim Zhang <qiangqing.zhang@nxp.com>; Mark Rutland
> <mark.rutland@arm.com>
> Cc: robin.murphy@arm.com; will@kernel.org;
> linux-arm-kernel@lists.infradead.org; dl-linux-imx <linux-imx@nxp.com>
> Subject: RE: [PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter support
> 
> 
> 
> > -----Original Message-----
> > From: Joakim Zhang
> > Sent: Monday, August 26, 2019 2:12 AM
> > To: Mark Rutland <mark.rutland@arm.com>
> > Cc: robin.murphy@arm.com; will@kernel.org; Frank Li
> > <frank.li@nxp.com>; linux-arm-kernel@lists.infradead.org; dl-linux-imx
> > <linux-imx@nxp.com>
> > Subject: RE: [PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter
> > support
> >
> >
> > > -----Original Message-----
> > > From: Mark Rutland <mark.rutland@arm.com>
> > > Sent: 2019年8月23日 20:57
> > > To: Joakim Zhang <qiangqing.zhang@nxp.com>
> > > Cc: robin.murphy@arm.com; will@kernel.org; Frank Li
> > > <frank.li@nxp.com>; linux-arm-kernel@lists.infradead.org;
> > > dl-linux-imx <linux-imx@nxp.com>
> > > Subject: Re: [PATCH V5 1/3] perf: imx8_ddr_perf: add AXI ID filter
> > > support
> > >
> > > On Thu, Aug 08, 2019 at 06:45:29AM +0000, Joakim Zhang wrote:
> > > > AXI filtering is used by CSV modes 0x41 and 0x42 to count reads or
> > > > writes with an ARID or AXID matching filter setting. Granularity
> > > > is at subsystem level. Implementation does not allow filtring
> > > > between masters within a subsystem. Filter is defined with 2
> configuration registers.
> > > >
> > > > --AXI_ID defines AxID matching value --AXI_MASKING defines which
> > > > bits of AxID are meaningful for the matching
> > > >
> > > > When non-masked bits are matching corresponding AXI_ID bits then
> > > > counter is incremented. This filter allows counting read or write
> > > > access from a subsystem or multiple subsystems.
> > > >
> > > > Perf counter is incremented if AxID && AXI_MASKING == AXI_ID &&
> > > > AXI_MASKING
> > >
> > > Just to check, the filter does not affect other events, right?
> >
> > [Joakim] Yes, this filter is only for events 0x41 and 0x42.
> >
> > > >
> > > > AXI_ID and AXI_MASKING are mapped on DPCR1 register in performance
> > > counter.
> > > >
> > > > Read and write AXI ID filter should write same value to DPCR1 if
> > > > want to specify at the same time as this filter is shared between
> counters.
> > > >
> > > > e.g.
> > > > perf stat -a -e
> > > >
> > >
> imx8_ddr0/axi-id-read,axi_id=0xMMMMDDDD/,imx8_ddr0/axi-id-write,axi_
> > > id
> > > > =0xMMMMDDDD/ cmd
> > > > MMMM: AXI_MASKING
> > > > DDDD: AXI_ID
> > >
> > > You might want to expose this to userspace as two fields:
> > >
> > > 	axi_id=DDDD
> > > 	axi_mask=MMMM
> > >
> > > ... where axi_mask is inverted (i.e. set bits are bits to ignore),
> > > so that the user can just specify axi_id to monitor a specific id,
> > > rather than having to specifiy axi_id=0xffff<id>.
> >
> > [Joakim] No, axi_mask is not inverted, should specify axi_id =
> > 0xffff<id> for the particular AXI ID. I will improve the commit message.
> > 		0: corresponding bit is masked
> > 		1: corresponding bit is not masked, i.e. used to do the matching
> >
> 
>  [Frank Li] Joakim,  mark's means is that revert it at your driver.
> Most case user just want to filter a AXID. For example 0x12
> 
> If you revert mask at driver,  user just need input axi_id = 0x12 instead of
> axi_id=0xffff0012.

[Joakim] Sorry for my misunderstanding, I will improve it in V7. Thanks.

> 
> > > >
> > > > ChangeLog:
> > > > V1 -> V2:
> > > > 	* add error log if user specifies read/write AXI ID filter at
> > > > 	the same time.
> > > > 	* of_device_get_match_data() instead of of_match_device(), and
> > > > 	remove the check of return value.
> > > > V2 -> V3:
> > > > 	* move the AXI ID check to event_add().
> > > > 	* add support for same value of axi_id.
> > > > V3 -> V4:
> > > > 	* move the AXI ID check to event_init().
> > > > V4 -> V5:
> > > > 	* reject event group if AXI ID not consistent in event_init().
> > > >
> > > > Signed-off-by: Joakim Zhang <qiangqing.zhang@nxp.com>
> > > > ---
> > > >  drivers/perf/fsl_imx8_ddr_perf.c | 63
> > > > +++++++++++++++++++++++++++++++-
> > > >  1 file changed, 61 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/drivers/perf/fsl_imx8_ddr_perf.c
> > > > b/drivers/perf/fsl_imx8_ddr_perf.c
> > > > index 63fe21600072..f25cf5cbe156 100644
> > > > --- a/drivers/perf/fsl_imx8_ddr_perf.c
> > > > +++ b/drivers/perf/fsl_imx8_ddr_perf.c
> > > > @@ -42,9 +42,22 @@
> > > >
> > > >  static DEFINE_IDA(ddr_ida);
> > > >
> > > > +/* DDR Perf hardware feature */
> > > > +#define DDR_CAP_AXI_ID_FILTER		0x1	/* support AXI ID filter
> > */
> > > > +
> > > > +struct fsl_ddr_devtype_data {
> > > > +	unsigned int quirks;	/* quirks needed for different DDR Perf core */
> > > > +};
> > > > +
> > > > +static const struct fsl_ddr_devtype_data imx8_devtype_data;
> > > > +
> > > > +static const struct fsl_ddr_devtype_data imx8m_devtype_data = {
> > > > +	.quirks = DDR_CAP_AXI_ID_FILTER, };
> > > > +
> > > >  static const struct of_device_id imx_ddr_pmu_dt_ids[] = {
> > > > -	{ .compatible = "fsl,imx8-ddr-pmu",},
> > > > -	{ .compatible = "fsl,imx8m-ddr-pmu",},
> > > > +	{ .compatible = "fsl,imx8-ddr-pmu", .data = &imx8_devtype_data},
> > > > +	{ .compatible = "fsl,imx8m-ddr-pmu", .data =
> > > > +&imx8m_devtype_data},
> > > >  	{ /* sentinel */ }
> > > >  };
> > >
> > > Are new DDR PMUs going to lack this feature?
> > >
> > > If all PMUs the driver supports have it, then we don't need this
> > > quirk/feature list.
> > >
> > > That would make the subsequent code a bit nicer, as all the
> > > filtering code would lose a level of indentation.
> >
> > [Joakim] This feature may drop, some coming SOCs will improve this AXI
> > ID filtering, let it can filter from different ID separately, and ang other
> extensions.
> >
> > > >
> > > > @@ -57,6 +70,7 @@ struct ddr_pmu {
> > > >  	struct perf_event *events[NUM_COUNTERS];
> > > >  	int active_events;
> > > >  	enum cpuhp_state cpuhp_state;
> > > > +	const struct fsl_ddr_devtype_data *devtype_data;
> > > >  	int irq;
> > > >  	int id;
> > > >  };
> > > > @@ -128,6 +142,8 @@ static struct attribute *ddr_perf_events_attrs[] =
> {
> > > >  	IMX8_DDR_PMU_EVENT_ATTR(refresh, 0x37),
> > > >  	IMX8_DDR_PMU_EVENT_ATTR(write, 0x38),
> > > >  	IMX8_DDR_PMU_EVENT_ATTR(raw-hazard, 0x39),
> > > > +	IMX8_DDR_PMU_EVENT_ATTR(axi-id-read, 0x41),
> > > > +	IMX8_DDR_PMU_EVENT_ATTR(axi-id-write, 0x42),
> > > >  	NULL,
> > > >  };
> > > >
> > > > @@ -137,9 +153,11 @@ static struct attribute_group
> > > > ddr_perf_events_attr_group = {  };
> > > >
> > > >  PMU_FORMAT_ATTR(event, "config:0-7");
> > > > +PMU_FORMAT_ATTR(axi_id, "config1:0-31");
> > > >
> > > >  static struct attribute *ddr_perf_format_attrs[] = {
> > > >  	&format_attr_event.attr,
> > > > +	&format_attr_axi_id.attr,
> > > >  	NULL,
> > > >  };
> > > >
> > > > @@ -189,6 +207,16 @@ static u32 ddr_perf_read_counter(struct
> > > > ddr_pmu
> > > *pmu, int counter)
> > > >  	return readl_relaxed(pmu->base + COUNTER_READ + counter *
> 4);  }
> > > >
> > > > +static bool ddr_perf_is_filtered(struct perf_event *event) {
> > > > +	return event->attr.config == 0x41 || event->attr.config == 0x42;
> > > > +}
> > > > +
> > > > +static u32 ddr_perf_filter_val(struct perf_event *event) {
> > > > +	return event->attr.config1;
> > > > +}
> > > > +
> > >
> > > Let's add another helper:
> > >
> > > static bool ddr_perf_filters_compatible(struct perf_event *a,
> > > 					struct perf_event *b)
> > > {
> > > 	if (!ddr_perf_is_filtered(a))
> > > 		return true;
> > > 	if (!ddr_perf_is_filtered(b))
> > > 		return true;
> > > 	return ddr_perf_filter_val(a) == ddr_perf_filter_val(b); }
> > >
> > > >  static int ddr_perf_event_init(struct perf_event *event)  {
> > > >  	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu); @@ -215,6
> +243,18
> > > @@
> > > > static int ddr_perf_event_init(struct perf_event *event)
> > > >  			!is_software_event(event->group_leader))
> > > >  		return -EINVAL;
> > > >
> > > > +	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> > > > +		bool is_filtered = ddr_perf_is_filtered(event);
> > > > +		u32 filter_val = ddr_perf_filter_val(event);
> > > > +
> > > > +		for_each_sibling_event(sibling, event->group_leader) {
> > > > +			if (is_filtered && ddr_perf_is_filtered(sibling) &&
> > > > +			    ddr_perf_filter_val(sibling) != filter_val) {
> > > > +				return -EINVAL;
> > > > +			}
> > > > +		}
> > > > +	}
> > >
> > > ... so this can be:
> > >
> > > 	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> > > 		if (!ddr_perf_filters_compatible(event, event->group_leader))
> > > 			return -EINVAL;
> > > 		for_each_sibling_event(sibling, event->group_leader) {
> > > 			if (!ddr_perf_filters_compatible(event, sibling))
> > > 				return -EINVAL;
> > > 		}
> > > 	}
> >
> > [Joakim] Got it.
> >
> > > Note: that checks group_leader, which you mised above. When the
> > > event is the group leader, it's trivially compatible with itself, so
> > > we don't need a special case there.
> > >
> > > > +
> > > >  	for_each_sibling_event(sibling, event->group_leader) {
> > > >  		if (sibling->pmu != event->pmu &&
> > > >  				!is_software_event(sibling))
> > > > @@ -288,6 +328,23 @@ static int ddr_perf_event_add(struct
> > > > perf_event
> > > *event, int flags)
> > > >  	int counter;
> > > >  	int cfg = event->attr.config;
> > > >
> > > > +	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> > > > +		int i;
> > > > +		bool is_filtered = ddr_perf_is_filtered(event);
> > > > +		u32 filter_val = ddr_perf_filter_val(event);
> > > > +
> > > > +		for (i = 1; i < NUM_COUNTERS; i++) {
> > > > +			if (is_filtered && pmu->events[i] &&
> > > > +			    ddr_perf_is_filtered(pmu->events[i]) &&
> > > > +			    ddr_perf_filter_val(pmu->events[i]) != filter_val) {
> > > > +				dev_dbg(pmu->dev, "Contradictory axi id filter
> > value\n");
> > > > +				return -EINVAL;
> > > > +			}
> > > > +		}
> > >
> > > ... and likewise:
> > >
> > > 	if (pmu->devtype_data->quirks & DDR_CAP_AXI_ID_FILTER) {
> > > 		int i;
> > >
> > > 		for (i = 1; i < NUM_COUNTERS; i++) {
> > > 			if (!ddr_perf_filters_compatible(event, pmu->events[i]))
> > > 				return -EINVAL;
> > > 		}
> > > 	}
> > >
> > > Please drop the dev_dbg() here, since when perf rotates events this
> > > can happen many times per second, and it's entirely legitimate.
> >
> > [Joakim] Got it.
> >
> > > Otherwise, this looks good to me.
> >
> > [Joakim] Thanks Mark, I will sent out a V6, please help review.
> >
> > Best Regards,
> > Joakim Zhang
> > > Thanks,
> > > Mark.
> > >
> > > > +
> > > > +		writel(filter_val, pmu->base + COUNTER_DPCR1);
> > > > +	}
> > > > +
> > > >  	counter = ddr_perf_alloc_counter(pmu, cfg);
> > > >  	if (counter < 0) {
> > > >  		dev_dbg(pmu->dev, "There are not enough counters\n"); @@ -
> > 472,6
> > > > +529,8 @@ static int ddr_perf_probe(struct platform_device *pdev)
> > > >  	if (!name)
> > > >  		return -ENOMEM;
> > > >
> > > > +	pmu->devtype_data = of_device_get_match_data(&pdev->dev);
> > > > +
> > > >  	pmu->cpu = raw_smp_processor_id();
> > > >  	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
> > > >  				      DDR_CPUHP_CB_NAME,
> > > > --
> > > > 2.17.1
> > > >
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel