diff mbox series

[1/3] dt-bindings: perf: fsl-imx-ddr: Add i.MX95 compatible

Message ID 20231127073208.1055466-1-xu.yang_2@nxp.com
State Changes Requested
Headers show
Series [1/3] dt-bindings: perf: fsl-imx-ddr: Add i.MX95 compatible | expand

Checks

Context Check Description
robh/checkpatch success
robh/patch-applied success
robh/dtbs-check warning build log
robh/dt-meta-schema success

Commit Message

Xu Yang Nov. 27, 2023, 7:32 a.m. UTC
i.MX95 has a more precise counting capability than i.MX93. This will add
a compatible for it.

Signed-off-by: Xu Yang <xu.yang_2@nxp.com>
---
 Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml | 1 +
 1 file changed, 1 insertion(+)

Comments

Frank Li Nov. 27, 2023, 3:25 p.m. UTC | #1
> -----Original Message-----
> From: Xu Yang <xu.yang_2@nxp.com>
> Sent: Monday, November 27, 2023 1:32 AM
> To: Frank Li <frank.li@nxp.com>; will@kernel.org; mark.rutland@arm.com;
> robh+dt@kernel.org; krzysztof.kozlowski+dt@linaro.org;
> conor+dt@kernel.org; shawnguo@kernel.org; s.hauer@pengutronix.de;
> kernel@pengutronix.de; festevam@gmail.com; irogers@google.com;
> namhyung@kernel.org; acme@kernel.org; john.g.garry@oracle.com
> Cc: james.clark@arm.com; mike.leach@linaro.org; leo.yan@linaro.org;
> peterz@infradead.org; mingo@redhat.com;
> alexander.shishkin@linux.intel.com; linux-arm-kernel@lists.infradead.org;
> devicetree@vger.kernel.org; linux-perf-users@vger.kernel.org; linux-
> kernel@vger.kernel.org
> Subject: [PATCH 2/3] perf: imx_perf: add support for i.MX95 platform
> 
> i.MX95 has a DDR PMU which is almostly same as i.MX93, it now supports
> read beat and write beat filter capabilities. This will add support for
> i.MX95 and enhance the driver to support specific filter handling for it.
> 
> Usage:
> 
> For read beat:
> ~# perf stat -a -I 1000 -e
> imx9_ddr0/eddrtq_pm_rd_beat_filt2,counter=3,axi_mask=ID_MASK,axi_id
> =ID/
> ~# perf stat -a -I 1000 -e
> imx9_ddr0/eddrtq_pm_rd_beat_filt1,counter=4,axi_mask=ID_MASK,axi_id
> =ID/
> ~# perf stat -a -I 1000 -e
> imx9_ddr0/eddrtq_pm_rd_beat_filt0,counter=5,axi_mask=ID_MASK,axi_id
> =ID/
> eg: For edma2: perf stat -a -I 1000 -e
> imx9_ddr0/eddrtq_pm_rd_beat_filt0,counter=5,axi_mask=0x00f,axi_id=0x0
> 0c/
> 
> For write beat:
> ~# perf stat -a -I 1000 -e
> imx9_ddr0/eddrtq_pm_wr_beat_filt,counter=2,axi_mask=ID_MASK,axi_id=
> ID/
> eg: For edma2: perf stat -a -I 1000 -e
> imx9_ddr0/eddrtq_pm_wr_beat_filt,counter=2,axi_mask=0x00f,axi_id=0x00
> c/
> 
> Signed-off-by: Xu Yang <xu.yang_2@nxp.com>
> ---
>  drivers/perf/fsl_imx9_ddr_perf.c | 187 +++++++++++++++++++++++++++--
> --
>  1 file changed, 164 insertions(+), 23 deletions(-)
> 
> diff --git a/drivers/perf/fsl_imx9_ddr_perf.c
> b/drivers/perf/fsl_imx9_ddr_perf.c
> index 5cf770a1bc31..5e531d94cf3b 100644
> --- a/drivers/perf/fsl_imx9_ddr_perf.c
> +++ b/drivers/perf/fsl_imx9_ddr_perf.c
> @@ -11,14 +11,24 @@
>  #include <linux/perf_event.h>
> 
>  /* Performance monitor configuration */
> -#define PMCFG1  			0x00
> -#define PMCFG1_RD_TRANS_FILT_EN 	BIT(31)
> -#define PMCFG1_WR_TRANS_FILT_EN 	BIT(30)
> -#define PMCFG1_RD_BT_FILT_EN 		BIT(29)
> -#define PMCFG1_ID_MASK  		GENMASK(17, 0)
> +#define PMCFG1				0x00
> +#define MX93_PMCFG1_RD_TRANS_FILT_EN	BIT(31)
> +#define MX93_PMCFG1_WR_TRANS_FILT_EN	BIT(30)
> +#define MX93_PMCFG1_RD_BT_FILT_EN	BIT(29)
> +#define MX93_PMCFG1_ID_MASK		GENMASK(17, 0)
> 
> -#define PMCFG2  			0x04
> -#define PMCFG2_ID			GENMASK(17, 0)
> +#define MX95_PMCFG1_WR_BEAT_FILT_EN	BIT(31)
> +#define MX95_PMCFG1_RD_BEAT_FILT_EN	BIT(30)
> +
> +#define PMCFG2				0x04
> +#define MX93_PMCFG2_ID			GENMASK(17, 0)
> +
> +#define PMCFG3				0x08
> +#define PMCFG4				0x0C
> +#define PMCFG5				0x10
> +#define PMCFG6				0x14
> +#define MX95_PMCFG_ID_MASK		GENMASK(9, 0)
> +#define MX95_PMCFG_ID			GENMASK(25, 16)
> 
>  /* Global control register affects all counters and takes priority over local
> control registers */
>  #define PMGC0		0x40
> @@ -71,12 +81,22 @@ static const struct imx_ddr_devtype_data
> imx93_devtype_data = {
>  	.identifier = "imx93",
>  };
> 
> +static const struct imx_ddr_devtype_data imx95_devtype_data = {
> +	.identifier = "imx95",
> +};
> +
>  static const struct of_device_id imx_ddr_pmu_dt_ids[] = {
>  	{.compatible = "fsl,imx93-ddr-pmu", .data = &imx93_devtype_data},
> +	{.compatible = "fsl,imx95-ddr-pmu", .data = &imx95_devtype_data},
>  	{ /* sentinel */ }
>  };
>  MODULE_DEVICE_TABLE(of, imx_ddr_pmu_dt_ids);
> 
> +static inline bool is_imx93(struct ddr_pmu *pmu)
> +{
> +	return pmu->devtype_data == &imx93_devtype_data;
> +}
> +
>  static ssize_t ddr_perf_identifier_show(struct device *dev,
>  					struct device_attribute *attr,
>  					char *page)
> @@ -178,7 +198,6 @@ static struct attribute *ddr_perf_events_attrs[] = {
>  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_ld_wiq_6, 70),
>  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_ld_wiq_7, 71),
>  	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pmon_empty, 72),
> -	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_trans_filt, 73),
> 
>  	/* counter3 specific events */
>  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_collision_0, 64),
> @@ -190,7 +209,6 @@ static struct attribute *ddr_perf_events_attrs[] = {
>  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_collision_6, 70),
>  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_collision_7, 71),
>  	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pmon_full, 72),
> -	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_wr_trans_filt, 73),
> 
>  	/* counter4 specific events */
>  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_open_0, 64),
> @@ -202,7 +220,6 @@ static struct attribute *ddr_perf_events_attrs[] = {
>  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_open_6, 70),
>  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_open_7, 71),
>  	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pmon_ld_rdq2_rmw, 72),
> -	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_beat_filt, 73),
> 
>  	/* counter5 specific events */
>  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_valid_start_0, 64),
> @@ -242,6 +259,26 @@ static const struct attribute_group
> ddr_perf_events_attr_group = {
>  	.attrs = ddr_perf_events_attrs,
>  };
> 
> +static struct attribute *imx93_ddr_perf_events_attrs[] = {
> +	/* counter2 specific events */
> +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_trans_filt, 73),
> +	/* counter3 specific events */
> +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_wr_trans_filt, 73),
> +	/* counter4 specific events */
> +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_beat_filt, 73),
> +};
> +
> +static struct attribute *imx95_ddr_perf_events_attrs[] = {
> +	/* counter2 specific events */
> +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_wr_beat_filt, 73),
> +	/* counter3 specific events */
> +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_beat_filt2, 73),
> +	/* counter4 specific events */
> +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_beat_filt1, 73),
> +	/* counter5 specific events */
> +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_beat_filt0, 73),
> +};
> +
>  PMU_FORMAT_ATTR(event, "config:0-7");
>  PMU_FORMAT_ATTR(counter, "config:8-15");
>  PMU_FORMAT_ATTR(axi_id, "config1:0-17");
> @@ -361,7 +398,7 @@ static void ddr_perf_counter_local_config(struct
> ddr_pmu *pmu, int config,
>  	}
>  }
> 
> -static void ddr_perf_monitor_config(struct ddr_pmu *pmu, int cfg, int cfg1,
> int cfg2)
> +static void imx93_ddr_perf_monitor_config(struct ddr_pmu *pmu, int cfg,
> int cfg1, int cfg2)
>  {
>  	u32 pmcfg1, pmcfg2;
>  	int event, counter;
> @@ -372,30 +409,80 @@ static void ddr_perf_monitor_config(struct
> ddr_pmu *pmu, int cfg, int cfg1, int
>  	pmcfg1 = readl_relaxed(pmu->base + PMCFG1);
> 
>  	if (counter == 2 && event == 73)
> -		pmcfg1 |= PMCFG1_RD_TRANS_FILT_EN;
> +		pmcfg1 |= MX93_PMCFG1_RD_TRANS_FILT_EN;
>  	else if (counter == 2 && event != 73)
> -		pmcfg1 &= ~PMCFG1_RD_TRANS_FILT_EN;
> +		pmcfg1 &= ~MX93_PMCFG1_RD_TRANS_FILT_EN;
> 
>  	if (counter == 3 && event == 73)
> -		pmcfg1 |= PMCFG1_WR_TRANS_FILT_EN;
> +		pmcfg1 |= MX93_PMCFG1_WR_TRANS_FILT_EN;
>  	else if (counter == 3 && event != 73)
> -		pmcfg1 &= ~PMCFG1_WR_TRANS_FILT_EN;
> +		pmcfg1 &= ~MX93_PMCFG1_WR_TRANS_FILT_EN;
> 
>  	if (counter == 4 && event == 73)
> -		pmcfg1 |= PMCFG1_RD_BT_FILT_EN;
> +		pmcfg1 |= MX93_PMCFG1_RD_BT_FILT_EN;
>  	else if (counter == 4 && event != 73)
> -		pmcfg1 &= ~PMCFG1_RD_BT_FILT_EN;
> +		pmcfg1 &= ~MX93_PMCFG1_RD_BT_FILT_EN;
> 
> -	pmcfg1 &= ~FIELD_PREP(PMCFG1_ID_MASK, 0x3FFFF);
> -	pmcfg1 |= FIELD_PREP(PMCFG1_ID_MASK, cfg2);
> +	pmcfg1 &= ~FIELD_PREP(MX93_PMCFG1_ID_MASK, 0x3FFFF);
> +	pmcfg1 |= FIELD_PREP(MX93_PMCFG1_ID_MASK, cfg2);
>  	writel(pmcfg1, pmu->base + PMCFG1);
> 
>  	pmcfg2 = readl_relaxed(pmu->base + PMCFG2);
> -	pmcfg2 &= ~FIELD_PREP(PMCFG2_ID, 0x3FFFF);
> -	pmcfg2 |= FIELD_PREP(PMCFG2_ID, cfg1);
> +	pmcfg2 &= ~FIELD_PREP(MX93_PMCFG2_ID, 0x3FFFF);
> +	pmcfg2 |= FIELD_PREP(MX93_PMCFG2_ID, cfg1);

Suggest created a sperate rename patch, which prepare for IMX95.

Frank

>  	writel(pmcfg2, pmu->base + PMCFG2);
>  }
> 
> +static void imx95_ddr_perf_monitor_config(struct ddr_pmu *pmu, int cfg,
> int cfg1, int cfg2)
> +{
> +	u32 pmcfg1, pmcfg, offset = 0;
> +	int event, counter;
> +
> +	event = cfg & 0x000000FF;
> +	counter = (cfg & 0x0000FF00) >> 8;
> +
> +	pmcfg1 = readl_relaxed(pmu->base + PMCFG1);
> +
> +	if (counter == 2 && event == 73) {
> +		pmcfg1 |= MX95_PMCFG1_WR_BEAT_FILT_EN;
> +		offset = PMCFG3;
> +	} else if (counter == 2 && event != 73) {
> +		pmcfg1 &= ~MX95_PMCFG1_WR_BEAT_FILT_EN;
> +	}
> +
> +	if (counter == 3 && event == 73) {
> +		pmcfg1 |= MX95_PMCFG1_RD_BEAT_FILT_EN;
> +		offset = PMCFG4;
> +	} else if (counter == 3 && event != 73) {
> +		pmcfg1 &= ~MX95_PMCFG1_RD_BEAT_FILT_EN;
> +	}
> +
> +	if (counter == 4 && event == 73) {
> +		pmcfg1 |= MX95_PMCFG1_RD_BEAT_FILT_EN;
> +		offset = PMCFG5;
> +	} else if (counter == 4 && event != 73) {
> +		pmcfg1 &= ~MX95_PMCFG1_RD_BEAT_FILT_EN;
> +	}
> +
> +	if (counter == 5 && event == 73) {
> +		pmcfg1 |= MX95_PMCFG1_RD_BEAT_FILT_EN;
> +		offset = PMCFG6;
> +	} else if (counter == 5 && event != 73) {
> +		pmcfg1 &= ~MX95_PMCFG1_RD_BEAT_FILT_EN;
> +	}
> +
> +	writel(pmcfg1, pmu->base + PMCFG1);
> +
> +	if (offset) {
> +		pmcfg = readl_relaxed(pmu->base + offset);
> +		pmcfg &= ~FIELD_PREP(MX95_PMCFG_ID_MASK, 0x3FF);
> +		pmcfg |= FIELD_PREP(MX95_PMCFG_ID_MASK, cfg2);
> +		pmcfg &= ~FIELD_PREP(MX95_PMCFG_ID, 0x3FF);

It should be reductant. 

> +		pmcfg |= FIELD_PREP(MX95_PMCFG_ID, cfg1);
> +		writel(pmcfg, pmu->base + offset);
> +	}
> +}
> +
>  static void ddr_perf_event_update(struct perf_event *event)
>  {
>  	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
> @@ -479,8 +566,13 @@ static int ddr_perf_event_add(struct perf_event
> *event, int flags)
>  	if (flags & PERF_EF_START)
>  		ddr_perf_event_start(event, flags);
> 
> -	/* read trans, write trans, read beat */
> -	ddr_perf_monitor_config(pmu, cfg, cfg1, cfg2);
> +	if (is_imx93(pmu)) {
> +		/* read trans, write trans, read beat */
> +		imx93_ddr_perf_monitor_config(pmu, cfg, cfg1, cfg2);
> +	} else {
> +		/* write beat, read beat2, read beat1, read beat */
> +		imx95_ddr_perf_monitor_config(pmu, cfg, cfg1, cfg2);
> +	}

Did you run checkpatch?  no "{" need?, not sure have comments case.
> 
>  	return 0;
>  }
> @@ -596,6 +688,49 @@ static int ddr_perf_offline_cpu(unsigned int cpu,
> struct hlist_node *node)
>  	return 0;
>  }
> 
> +static int ddr_perf_add_events(struct ddr_pmu *pmu)
> +{
> +	int i, ret, events;
> +	struct attribute **attrs;
> +	struct device *pmu_dev = pmu->pmu.dev;
> +
> +	if (is_imx93(pmu)) {
> +		events = sizeof(imx93_ddr_perf_events_attrs)/sizeof(struct
> attribute *);
> +		attrs = imx93_ddr_perf_events_attrs;
> +	} else {
> +		events = sizeof(imx95_ddr_perf_events_attrs)/sizeof(struct
> attribute *);
> +		attrs = imx95_ddr_perf_events_attrs;
> +	}

Can you put "attrs" and "events" into drvdata?

> +
> +	for (i = 0; i < events; i++) {
> +		ret = sysfs_add_file_to_group(&pmu_dev->kobj, attrs[i],
> "events");
> +		if (ret) {
> +			dev_warn(pmu->dev, "i.MX9 DDR Perf add events
> failed (%d)\n", ret);
> +			return ret;
> +		}
> +	}
> +
> +	return 0;
> +}
> +
> +static void ddr_perf_remove_events(struct ddr_pmu *pmu)
> +{
> +	int i, events;
> +	struct attribute **attrs;
> +	struct device *pmu_dev = pmu->pmu.dev;
> +
> +	if (is_imx93(pmu)) {
> +		events = sizeof(imx93_ddr_perf_events_attrs)/sizeof(struct
> attribute *);
> +		attrs = imx93_ddr_perf_events_attrs;
> +	} else {
> +		events = sizeof(imx95_ddr_perf_events_attrs)/sizeof(struct
> attribute *);
> +		attrs = imx95_ddr_perf_events_attrs;
> +	}

Can you put "attrs" and "events" into drvdata?

> +
> +	for (i = 0; i < events; i++)
> +		sysfs_remove_file_from_group(&pmu_dev->kobj, attrs[i],
> "events");
> +}
> +
>  static int ddr_perf_probe(struct platform_device *pdev)
>  {
>  	struct ddr_pmu *pmu;
> @@ -666,6 +801,10 @@ static int ddr_perf_probe(struct platform_device
> *pdev)
>  	if (ret)
>  		goto ddr_perf_err;
> 
> +	ret = ddr_perf_add_events(pmu);
> +	if (ret)
> +		dev_warn(&pdev->dev, "i.MX9 DDR Perf filter events are
> missing\n");
> +
>  	return 0;
> 
>  ddr_perf_err:
> @@ -683,6 +822,8 @@ static int ddr_perf_remove(struct platform_device
> *pdev)
>  {
>  	struct ddr_pmu *pmu = platform_get_drvdata(pdev);
> 
> +	ddr_perf_remove_events(pmu);
> +
>  	cpuhp_state_remove_instance_nocalls(pmu->cpuhp_state, &pmu-
> >node);
>  	cpuhp_remove_multi_state(pmu->cpuhp_state);
> 
> --
> 2.34.1
Conor Dooley Nov. 27, 2023, 5:36 p.m. UTC | #2
On Mon, Nov 27, 2023 at 03:32:06PM +0800, Xu Yang wrote:
> i.MX95 has a more precise counting capability than i.MX93. This will add
> a compatible for it.

It is hard to tell from this comment, but I figure this "more precise
capability" is not an option you can enable, but instead makes the
programming model of this device different to that of the imx93?

Thanks,
Conor.

> 
> Signed-off-by: Xu Yang <xu.yang_2@nxp.com>
> ---
>  Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml b/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml
> index e9fad4b3de68..1bc7bf1c8368 100644
> --- a/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml
> +++ b/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml
> @@ -20,6 +20,7 @@ properties:
>            - fsl,imx8mn-ddr-pmu
>            - fsl,imx8mp-ddr-pmu
>            - fsl,imx93-ddr-pmu
> +          - fsl,imx95-ddr-pmu
>        - items:
>            - enum:
>                - fsl,imx8mm-ddr-pmu
> -- 
> 2.34.1
>
Xu Yang Dec. 1, 2023, 6:41 a.m. UTC | #3
Hi Conor,

> 
> On Mon, Nov 27, 2023 at 03:32:06PM +0800, Xu Yang wrote:
> > i.MX95 has a more precise counting capability than i.MX93. This will add
> > a compatible for it.
> 
> It is hard to tell from this comment, but I figure this "more precise
> capability" is not an option you can enable, but instead makes the
> programming model of this device different to that of the imx93?

Actually, imx95 is compatible with imx93 except AXI ID filter capability.
But for AXI ID filter, imx95 is using different registers and bits from
imx93 for filter configuration. To distinguish them, I need use different
compatible because programming model cannot recognize which device is
running.

compatible = "fsl,imx95-ddr-pmu";
compatible = "fsl,imx95-ddr-pmu", "fsl,imx93-ddr-pmu";

Both above compatible is okay for me. Therefore, "fsl,imx95-ddr-pmu" is needed.

Thanks,
Xu Yang

> 
> Thanks,
> Conor.
> 
> >
> > Signed-off-by: Xu Yang <xu.yang_2@nxp.com>
> > ---
> >  Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml | 1 +
> >  1 file changed, 1 insertion(+)
> >
> > diff --git a/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml b/Documentation/devicetree/bindings/perf/fsl-
> imx-ddr.yaml
> > index e9fad4b3de68..1bc7bf1c8368 100644
> > --- a/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml
> > +++ b/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml
> > @@ -20,6 +20,7 @@ properties:
> >            - fsl,imx8mn-ddr-pmu
> >            - fsl,imx8mp-ddr-pmu
> >            - fsl,imx93-ddr-pmu
> > +          - fsl,imx95-ddr-pmu
> >        - items:
> >            - enum:
> >                - fsl,imx8mm-ddr-pmu
> > --
> > 2.34.1
> >
Xu Yang Dec. 1, 2023, 8:56 a.m. UTC | #4
Hi Frank,

> -----Original Message-----
> From: Frank Li <frank.li@nxp.com>
> Sent: Monday, November 27, 2023 11:25 PM
> To: Xu Yang <xu.yang_2@nxp.com>; will@kernel.org; mark.rutland@arm.com; robh+dt@kernel.org;
> krzysztof.kozlowski+dt@linaro.org; conor+dt@kernel.org; shawnguo@kernel.org; s.hauer@pengutronix.de;
> kernel@pengutronix.de; festevam@gmail.com; irogers@google.com; namhyung@kernel.org; acme@kernel.org;
> john.g.garry@oracle.com
> Cc: james.clark@arm.com; mike.leach@linaro.org; leo.yan@linaro.org; peterz@infradead.org; mingo@redhat.com;
> alexander.shishkin@linux.intel.com; linux-arm-kernel@lists.infradead.org; devicetree@vger.kernel.org; linux-perf-
> users@vger.kernel.org; linux-kernel@vger.kernel.org
> Subject: RE: [PATCH 2/3] perf: imx_perf: add support for i.MX95 platform
> 
> 
> 
> > -----Original Message-----
> > From: Xu Yang <xu.yang_2@nxp.com>
> > Sent: Monday, November 27, 2023 1:32 AM
> > To: Frank Li <frank.li@nxp.com>; will@kernel.org; mark.rutland@arm.com;
> > robh+dt@kernel.org; krzysztof.kozlowski+dt@linaro.org;
> > conor+dt@kernel.org; shawnguo@kernel.org; s.hauer@pengutronix.de;
> > kernel@pengutronix.de; festevam@gmail.com; irogers@google.com;
> > namhyung@kernel.org; acme@kernel.org; john.g.garry@oracle.com
> > Cc: james.clark@arm.com; mike.leach@linaro.org; leo.yan@linaro.org;
> > peterz@infradead.org; mingo@redhat.com;
> > alexander.shishkin@linux.intel.com; linux-arm-kernel@lists.infradead.org;
> > devicetree@vger.kernel.org; linux-perf-users@vger.kernel.org; linux-
> > kernel@vger.kernel.org
> > Subject: [PATCH 2/3] perf: imx_perf: add support for i.MX95 platform
> >
> > i.MX95 has a DDR PMU which is almostly same as i.MX93, it now supports
> > read beat and write beat filter capabilities. This will add support for
> > i.MX95 and enhance the driver to support specific filter handling for it.
> >
> > Usage:
> >
> > For read beat:
> > ~# perf stat -a -I 1000 -e
> > imx9_ddr0/eddrtq_pm_rd_beat_filt2,counter=3,axi_mask=ID_MASK,axi_id
> > =ID/
> > ~# perf stat -a -I 1000 -e
> > imx9_ddr0/eddrtq_pm_rd_beat_filt1,counter=4,axi_mask=ID_MASK,axi_id
> > =ID/
> > ~# perf stat -a -I 1000 -e
> > imx9_ddr0/eddrtq_pm_rd_beat_filt0,counter=5,axi_mask=ID_MASK,axi_id
> > =ID/
> > eg: For edma2: perf stat -a -I 1000 -e
> > imx9_ddr0/eddrtq_pm_rd_beat_filt0,counter=5,axi_mask=0x00f,axi_id=0x0
> > 0c/
> >
> > For write beat:
> > ~# perf stat -a -I 1000 -e
> > imx9_ddr0/eddrtq_pm_wr_beat_filt,counter=2,axi_mask=ID_MASK,axi_id=
> > ID/
> > eg: For edma2: perf stat -a -I 1000 -e
> > imx9_ddr0/eddrtq_pm_wr_beat_filt,counter=2,axi_mask=0x00f,axi_id=0x00
> > c/
> >
> > Signed-off-by: Xu Yang <xu.yang_2@nxp.com>
> > ---
> >  drivers/perf/fsl_imx9_ddr_perf.c | 187 +++++++++++++++++++++++++++--
> > --
> >  1 file changed, 164 insertions(+), 23 deletions(-)
> >
> > diff --git a/drivers/perf/fsl_imx9_ddr_perf.c
> > b/drivers/perf/fsl_imx9_ddr_perf.c
> > index 5cf770a1bc31..5e531d94cf3b 100644
> > --- a/drivers/perf/fsl_imx9_ddr_perf.c
> > +++ b/drivers/perf/fsl_imx9_ddr_perf.c
> > @@ -11,14 +11,24 @@
> >  #include <linux/perf_event.h>
> >
> >  /* Performance monitor configuration */
> > -#define PMCFG1  			0x00
> > -#define PMCFG1_RD_TRANS_FILT_EN 	BIT(31)
> > -#define PMCFG1_WR_TRANS_FILT_EN 	BIT(30)
> > -#define PMCFG1_RD_BT_FILT_EN 		BIT(29)
> > -#define PMCFG1_ID_MASK  		GENMASK(17, 0)
> > +#define PMCFG1				0x00
> > +#define MX93_PMCFG1_RD_TRANS_FILT_EN	BIT(31)
> > +#define MX93_PMCFG1_WR_TRANS_FILT_EN	BIT(30)
> > +#define MX93_PMCFG1_RD_BT_FILT_EN	BIT(29)
> > +#define MX93_PMCFG1_ID_MASK		GENMASK(17, 0)
> >
> > -#define PMCFG2  			0x04
> > -#define PMCFG2_ID			GENMASK(17, 0)
> > +#define MX95_PMCFG1_WR_BEAT_FILT_EN	BIT(31)
> > +#define MX95_PMCFG1_RD_BEAT_FILT_EN	BIT(30)
> > +
> > +#define PMCFG2				0x04
> > +#define MX93_PMCFG2_ID			GENMASK(17, 0)
> > +
> > +#define PMCFG3				0x08
> > +#define PMCFG4				0x0C
> > +#define PMCFG5				0x10
> > +#define PMCFG6				0x14
> > +#define MX95_PMCFG_ID_MASK		GENMASK(9, 0)
> > +#define MX95_PMCFG_ID			GENMASK(25, 16)
> >
> >  /* Global control register affects all counters and takes priority over local
> > control registers */
> >  #define PMGC0		0x40
> > @@ -71,12 +81,22 @@ static const struct imx_ddr_devtype_data
> > imx93_devtype_data = {
> >  	.identifier = "imx93",
> >  };
> >
> > +static const struct imx_ddr_devtype_data imx95_devtype_data = {
> > +	.identifier = "imx95",
> > +};
> > +
> >  static const struct of_device_id imx_ddr_pmu_dt_ids[] = {
> >  	{.compatible = "fsl,imx93-ddr-pmu", .data = &imx93_devtype_data},
> > +	{.compatible = "fsl,imx95-ddr-pmu", .data = &imx95_devtype_data},
> >  	{ /* sentinel */ }
> >  };
> >  MODULE_DEVICE_TABLE(of, imx_ddr_pmu_dt_ids);
> >
> > +static inline bool is_imx93(struct ddr_pmu *pmu)
> > +{
> > +	return pmu->devtype_data == &imx93_devtype_data;
> > +}
> > +
> >  static ssize_t ddr_perf_identifier_show(struct device *dev,
> >  					struct device_attribute *attr,
> >  					char *page)
> > @@ -178,7 +198,6 @@ static struct attribute *ddr_perf_events_attrs[] = {
> >  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_ld_wiq_6, 70),
> >  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_ld_wiq_7, 71),
> >  	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pmon_empty, 72),
> > -	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_trans_filt, 73),
> >
> >  	/* counter3 specific events */
> >  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_collision_0, 64),
> > @@ -190,7 +209,6 @@ static struct attribute *ddr_perf_events_attrs[] = {
> >  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_collision_6, 70),
> >  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_collision_7, 71),
> >  	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pmon_full, 72),
> > -	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_wr_trans_filt, 73),
> >
> >  	/* counter4 specific events */
> >  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_open_0, 64),
> > @@ -202,7 +220,6 @@ static struct attribute *ddr_perf_events_attrs[] = {
> >  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_open_6, 70),
> >  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_row_open_7, 71),
> >  	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pmon_ld_rdq2_rmw, 72),
> > -	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_beat_filt, 73),
> >
> >  	/* counter5 specific events */
> >  	IMX9_DDR_PMU_EVENT_ATTR(ddrc_qx_valid_start_0, 64),
> > @@ -242,6 +259,26 @@ static const struct attribute_group
> > ddr_perf_events_attr_group = {
> >  	.attrs = ddr_perf_events_attrs,
> >  };
> >
> > +static struct attribute *imx93_ddr_perf_events_attrs[] = {
> > +	/* counter2 specific events */
> > +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_trans_filt, 73),
> > +	/* counter3 specific events */
> > +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_wr_trans_filt, 73),
> > +	/* counter4 specific events */
> > +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_beat_filt, 73),
> > +};
> > +
> > +static struct attribute *imx95_ddr_perf_events_attrs[] = {
> > +	/* counter2 specific events */
> > +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_wr_beat_filt, 73),
> > +	/* counter3 specific events */
> > +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_beat_filt2, 73),
> > +	/* counter4 specific events */
> > +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_beat_filt1, 73),
> > +	/* counter5 specific events */
> > +	IMX9_DDR_PMU_EVENT_ATTR(eddrtq_pm_rd_beat_filt0, 73),
> > +};
> > +
> >  PMU_FORMAT_ATTR(event, "config:0-7");
> >  PMU_FORMAT_ATTR(counter, "config:8-15");
> >  PMU_FORMAT_ATTR(axi_id, "config1:0-17");
> > @@ -361,7 +398,7 @@ static void ddr_perf_counter_local_config(struct
> > ddr_pmu *pmu, int config,
> >  	}
> >  }
> >
> > -static void ddr_perf_monitor_config(struct ddr_pmu *pmu, int cfg, int cfg1,
> > int cfg2)
> > +static void imx93_ddr_perf_monitor_config(struct ddr_pmu *pmu, int cfg,
> > int cfg1, int cfg2)
> >  {
> >  	u32 pmcfg1, pmcfg2;
> >  	int event, counter;
> > @@ -372,30 +409,80 @@ static void ddr_perf_monitor_config(struct
> > ddr_pmu *pmu, int cfg, int cfg1, int
> >  	pmcfg1 = readl_relaxed(pmu->base + PMCFG1);
> >
> >  	if (counter == 2 && event == 73)
> > -		pmcfg1 |= PMCFG1_RD_TRANS_FILT_EN;
> > +		pmcfg1 |= MX93_PMCFG1_RD_TRANS_FILT_EN;
> >  	else if (counter == 2 && event != 73)
> > -		pmcfg1 &= ~PMCFG1_RD_TRANS_FILT_EN;
> > +		pmcfg1 &= ~MX93_PMCFG1_RD_TRANS_FILT_EN;
> >
> >  	if (counter == 3 && event == 73)
> > -		pmcfg1 |= PMCFG1_WR_TRANS_FILT_EN;
> > +		pmcfg1 |= MX93_PMCFG1_WR_TRANS_FILT_EN;
> >  	else if (counter == 3 && event != 73)
> > -		pmcfg1 &= ~PMCFG1_WR_TRANS_FILT_EN;
> > +		pmcfg1 &= ~MX93_PMCFG1_WR_TRANS_FILT_EN;
> >
> >  	if (counter == 4 && event == 73)
> > -		pmcfg1 |= PMCFG1_RD_BT_FILT_EN;
> > +		pmcfg1 |= MX93_PMCFG1_RD_BT_FILT_EN;
> >  	else if (counter == 4 && event != 73)
> > -		pmcfg1 &= ~PMCFG1_RD_BT_FILT_EN;
> > +		pmcfg1 &= ~MX93_PMCFG1_RD_BT_FILT_EN;
> >
> > -	pmcfg1 &= ~FIELD_PREP(PMCFG1_ID_MASK, 0x3FFFF);
> > -	pmcfg1 |= FIELD_PREP(PMCFG1_ID_MASK, cfg2);
> > +	pmcfg1 &= ~FIELD_PREP(MX93_PMCFG1_ID_MASK, 0x3FFFF);
> > +	pmcfg1 |= FIELD_PREP(MX93_PMCFG1_ID_MASK, cfg2);
> >  	writel(pmcfg1, pmu->base + PMCFG1);
> >
> >  	pmcfg2 = readl_relaxed(pmu->base + PMCFG2);
> > -	pmcfg2 &= ~FIELD_PREP(PMCFG2_ID, 0x3FFFF);
> > -	pmcfg2 |= FIELD_PREP(PMCFG2_ID, cfg1);
> > +	pmcfg2 &= ~FIELD_PREP(MX93_PMCFG2_ID, 0x3FFFF);
> > +	pmcfg2 |= FIELD_PREP(MX93_PMCFG2_ID, cfg1);
> 
> Suggest created a sperate rename patch, which prepare for IMX95.
> 
> Frank
> 
> >  	writel(pmcfg2, pmu->base + PMCFG2);
> >  }
> >
> > +static void imx95_ddr_perf_monitor_config(struct ddr_pmu *pmu, int cfg,
> > int cfg1, int cfg2)
> > +{
> > +	u32 pmcfg1, pmcfg, offset = 0;
> > +	int event, counter;
> > +
> > +	event = cfg & 0x000000FF;
> > +	counter = (cfg & 0x0000FF00) >> 8;
> > +
> > +	pmcfg1 = readl_relaxed(pmu->base + PMCFG1);
> > +
> > +	if (counter == 2 && event == 73) {
> > +		pmcfg1 |= MX95_PMCFG1_WR_BEAT_FILT_EN;
> > +		offset = PMCFG3;
> > +	} else if (counter == 2 && event != 73) {
> > +		pmcfg1 &= ~MX95_PMCFG1_WR_BEAT_FILT_EN;
> > +	}
> > +
> > +	if (counter == 3 && event == 73) {
> > +		pmcfg1 |= MX95_PMCFG1_RD_BEAT_FILT_EN;
> > +		offset = PMCFG4;
> > +	} else if (counter == 3 && event != 73) {
> > +		pmcfg1 &= ~MX95_PMCFG1_RD_BEAT_FILT_EN;
> > +	}
> > +
> > +	if (counter == 4 && event == 73) {
> > +		pmcfg1 |= MX95_PMCFG1_RD_BEAT_FILT_EN;
> > +		offset = PMCFG5;
> > +	} else if (counter == 4 && event != 73) {
> > +		pmcfg1 &= ~MX95_PMCFG1_RD_BEAT_FILT_EN;
> > +	}
> > +
> > +	if (counter == 5 && event == 73) {
> > +		pmcfg1 |= MX95_PMCFG1_RD_BEAT_FILT_EN;
> > +		offset = PMCFG6;
> > +	} else if (counter == 5 && event != 73) {
> > +		pmcfg1 &= ~MX95_PMCFG1_RD_BEAT_FILT_EN;
> > +	}
> > +
> > +	writel(pmcfg1, pmu->base + PMCFG1);
> > +
> > +	if (offset) {
> > +		pmcfg = readl_relaxed(pmu->base + offset);
> > +		pmcfg &= ~FIELD_PREP(MX95_PMCFG_ID_MASK, 0x3FF);
> > +		pmcfg |= FIELD_PREP(MX95_PMCFG_ID_MASK, cfg2);
> > +		pmcfg &= ~FIELD_PREP(MX95_PMCFG_ID, 0x3FF);
> 
> It should be reductant.

Okay, will change this.

> 
> > +		pmcfg |= FIELD_PREP(MX95_PMCFG_ID, cfg1);
> > +		writel(pmcfg, pmu->base + offset);
> > +	}
> > +}
> > +
> >  static void ddr_perf_event_update(struct perf_event *event)
> >  {
> >  	struct ddr_pmu *pmu = to_ddr_pmu(event->pmu);
> > @@ -479,8 +566,13 @@ static int ddr_perf_event_add(struct perf_event
> > *event, int flags)
> >  	if (flags & PERF_EF_START)
> >  		ddr_perf_event_start(event, flags);
> >
> > -	/* read trans, write trans, read beat */
> > -	ddr_perf_monitor_config(pmu, cfg, cfg1, cfg2);
> > +	if (is_imx93(pmu)) {
> > +		/* read trans, write trans, read beat */
> > +		imx93_ddr_perf_monitor_config(pmu, cfg, cfg1, cfg2);
> > +	} else {
> > +		/* write beat, read beat2, read beat1, read beat */
> > +		imx95_ddr_perf_monitor_config(pmu, cfg, cfg1, cfg2);
> > +	}
> 
> Did you run checkpatch?  no "{" need?, not sure have comments case.

No warning about this when run checkpatch.

> >
> >  	return 0;
> >  }
> > @@ -596,6 +688,49 @@ static int ddr_perf_offline_cpu(unsigned int cpu,
> > struct hlist_node *node)
> >  	return 0;
> >  }
> >
> > +static int ddr_perf_add_events(struct ddr_pmu *pmu)
> > +{
> > +	int i, ret, events;
> > +	struct attribute **attrs;
> > +	struct device *pmu_dev = pmu->pmu.dev;
> > +
> > +	if (is_imx93(pmu)) {
> > +		events = sizeof(imx93_ddr_perf_events_attrs)/sizeof(struct
> > attribute *);
> > +		attrs = imx93_ddr_perf_events_attrs;
> > +	} else {
> > +		events = sizeof(imx95_ddr_perf_events_attrs)/sizeof(struct
> > attribute *);
> > +		attrs = imx95_ddr_perf_events_attrs;
> > +	}
> 
> Can you put "attrs" and "events" into drvdata?

Okay. Will try it.

> 
> > +
> > +	for (i = 0; i < events; i++) {
> > +		ret = sysfs_add_file_to_group(&pmu_dev->kobj, attrs[i],
> > "events");
> > +		if (ret) {
> > +			dev_warn(pmu->dev, "i.MX9 DDR Perf add events
> > failed (%d)\n", ret);
> > +			return ret;
> > +		}
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +static void ddr_perf_remove_events(struct ddr_pmu *pmu)
> > +{
> > +	int i, events;
> > +	struct attribute **attrs;
> > +	struct device *pmu_dev = pmu->pmu.dev;
> > +
> > +	if (is_imx93(pmu)) {
> > +		events = sizeof(imx93_ddr_perf_events_attrs)/sizeof(struct
> > attribute *);
> > +		attrs = imx93_ddr_perf_events_attrs;
> > +	} else {
> > +		events = sizeof(imx95_ddr_perf_events_attrs)/sizeof(struct
> > attribute *);
> > +		attrs = imx95_ddr_perf_events_attrs;
> > +	}
> 
> Can you put "attrs" and "events" into drvdata?

Okay.

Thanks,
Xu Yang

> 
> > +
> > +	for (i = 0; i < events; i++)
> > +		sysfs_remove_file_from_group(&pmu_dev->kobj, attrs[i],
> > "events");
> > +}
> > +
> >  static int ddr_perf_probe(struct platform_device *pdev)
> >  {
> >  	struct ddr_pmu *pmu;
> > @@ -666,6 +801,10 @@ static int ddr_perf_probe(struct platform_device
> > *pdev)
> >  	if (ret)
> >  		goto ddr_perf_err;
> >
> > +	ret = ddr_perf_add_events(pmu);
> > +	if (ret)
> > +		dev_warn(&pdev->dev, "i.MX9 DDR Perf filter events are
> > missing\n");
> > +
> >  	return 0;
> >
> >  ddr_perf_err:
> > @@ -683,6 +822,8 @@ static int ddr_perf_remove(struct platform_device
> > *pdev)
> >  {
> >  	struct ddr_pmu *pmu = platform_get_drvdata(pdev);
> >
> > +	ddr_perf_remove_events(pmu);
> > +
> >  	cpuhp_state_remove_instance_nocalls(pmu->cpuhp_state, &pmu-
> > >node);
> >  	cpuhp_remove_multi_state(pmu->cpuhp_state);
> >
> > --
> > 2.34.1
Conor Dooley Dec. 1, 2023, 4:16 p.m. UTC | #5
On Fri, Dec 01, 2023 at 06:41:59AM +0000, Xu Yang wrote:
> Hi Conor,
> 
> > 
> > On Mon, Nov 27, 2023 at 03:32:06PM +0800, Xu Yang wrote:
> > > i.MX95 has a more precise counting capability than i.MX93. This will add
> > > a compatible for it.
> > 
> > It is hard to tell from this comment, but I figure this "more precise
> > capability" is not an option you can enable, but instead makes the
> > programming model of this device different to that of the imx93?
> 
> Actually, imx95 is compatible with imx93 except AXI ID filter capability.
> But for AXI ID filter, imx95 is using different registers and bits from
> imx93 for filter configuration.

This sounds like it conflicts with...

> To distinguish them, I need use different
> compatible because programming model cannot recognize which device is
> running.
> 
> compatible = "fsl,imx95-ddr-pmu";

> compatible = "fsl,imx95-ddr-pmu", "fsl,imx93-ddr-pmu";

...this. If drivers for the imx93 need changes to work on the imx95,
then `compatible = "fsl,imx95-ddr-pmu", "fsl,imx93-ddr-pmu";` cannot be
used. If they will work, with only the new imx95 features being
non-functional, then you can use it.

> Both above compatible is okay for me. Therefore, "fsl,imx95-ddr-pmu" is needed.
Xu Yang Dec. 6, 2023, 1:42 a.m. UTC | #6
Hi Hi Conor,

> 
> On Fri, Dec 01, 2023 at 06:41:59AM +0000, Xu Yang wrote:
> > Hi Conor,
> >
> > >
> > > On Mon, Nov 27, 2023 at 03:32:06PM +0800, Xu Yang wrote:
> > > > i.MX95 has a more precise counting capability than i.MX93. This will add
> > > > a compatible for it.
> > >
> > > It is hard to tell from this comment, but I figure this "more precise
> > > capability" is not an option you can enable, but instead makes the
> > > programming model of this device different to that of the imx93?
> >
> > Actually, imx95 is compatible with imx93 except AXI ID filter capability.
> > But for AXI ID filter, imx95 is using different registers and bits from
> > imx93 for filter configuration.
> 
> This sounds like it conflicts with...
> 
> > To distinguish them, I need use different
> > compatible because programming model cannot recognize which device is
> > running.
> >
> > compatible = "fsl,imx95-ddr-pmu";
> 
> > compatible = "fsl,imx95-ddr-pmu", "fsl,imx93-ddr-pmu";
> 
> ...this. If drivers for the imx93 need changes to work on the imx95,
> then `compatible = "fsl,imx95-ddr-pmu", "fsl,imx93-ddr-pmu";` cannot be
> used. If they will work, with only the new imx95 features being
> non-functional, then you can use it.

Yes, it is. When compatible = "fsl,imx95-ddr-pmu", "fsl,imx93-ddr-pmu" is used,
only the new imx95 features is not functional, other basic function works for
both imx95 and imx93.

Thanks,
Xu Yang

> 
> > Both above compatible is okay for me. Therefore, "fsl,imx95-ddr-pmu" is needed.
diff mbox series

Patch

diff --git a/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml b/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml
index e9fad4b3de68..1bc7bf1c8368 100644
--- a/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml
+++ b/Documentation/devicetree/bindings/perf/fsl-imx-ddr.yaml
@@ -20,6 +20,7 @@  properties:
           - fsl,imx8mn-ddr-pmu
           - fsl,imx8mp-ddr-pmu
           - fsl,imx93-ddr-pmu
+          - fsl,imx95-ddr-pmu
       - items:
           - enum:
               - fsl,imx8mm-ddr-pmu