[EXT] Re: [PATCH v4 3/6] perf: imx_perf: add support for i.MX95 platform
Xu Yang
xu.yang_2 at nxp.com
Thu Feb 29 02:27:03 PST 2024
Hi Will,
>
> On Wed, Jan 31, 2024 at 01:58:08PM +0800, Xu Yang wrote:
> > 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=0x00c/
> >
> > 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=0x00c/
> >
> > Signed-off-by: Xu Yang <xu.yang_2 at nxp.com>
> >
> > ---
> > Changes in v2:
> > - put soc spefific axi filter events to drvdata according
> > to franks suggestions.
> > - adjust pmcfg axi_id and axi_mask config
> > Changes in v3:
> > - no changes
> > Changes in v4:
> > - only contain imx95 parts
> > ---
> > drivers/perf/fsl_imx9_ddr_perf.c | 86 +++++++++++++++++++++++++++++++-
> > 1 file changed, 84 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/perf/fsl_imx9_ddr_perf.c b/drivers/perf/fsl_imx9_ddr_perf.c
> > index b1a58e9e1617..85aaaef7212f 100644
> > --- a/drivers/perf/fsl_imx9_ddr_perf.c
> > +++ b/drivers/perf/fsl_imx9_ddr_perf.c
> > @@ -17,9 +17,19 @@
> > #define MX93_PMCFG1_RD_BT_FILT_EN BIT(29)
> > #define MX93_PMCFG1_ID_MASK 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
> > /* Global control register bits */
> > @@ -240,6 +250,18 @@ static struct attribute *imx93_ddr_perf_events_attrs[] = {
> > NULL,
> > };
> >
> > +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),
> > + NULL,
> > +};
> > +
> > PMU_FORMAT_ATTR(event, "config:0-7");
> > PMU_FORMAT_ATTR(counter, "config:8-15");
> > PMU_FORMAT_ATTR(axi_id, "config1:0-17");
> > @@ -271,8 +293,14 @@ static const struct imx_ddr_devtype_data imx93_devtype_data = {
> > .attrs = imx93_ddr_perf_events_attrs,
> > };
> >
> > +static const struct imx_ddr_devtype_data imx95_devtype_data = {
> > + .identifier = "imx95",
> > + .attrs = imx95_ddr_perf_events_attrs,
> > +};
> > +
> > 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);
> > @@ -410,6 +438,56 @@ static void imx93_ddr_perf_monitor_config(struct ddr_pmu *pmu, int cfg, int cfg1
> > 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;
> > + }
>
> I think this would be much easier to read if you rewrote it as:
>
> switch (counter) {
> case 2:
> ...
> break;
> case 3:
>
>
> etc.
Okay, I will improve this.
>
> > + writel(pmcfg1, pmu->base + PMCFG1);
>
> writel_relaxed()? (since you're already using readl_relaxed()).
Okay.
>
> > +
> > + if (offset) {
> > + pmcfg = readl_relaxed(pmu->base + offset);
> > + pmcfg &= ~(FIELD_PREP(MX95_PMCFG_ID_MASK, 0x3FF) |
> > + FIELD_PREP(MX95_PMCFG_ID, 0x3FF));
> > + pmcfg |= (FIELD_PREP(MX95_PMCFG_ID_MASK, cfg2) |
> > + FIELD_PREP(MX95_PMCFG_ID, cfg1));
> > + writel(pmcfg, pmu->base + offset);
>
> Similarly.
Okay.
Thanks,
Xu Yang
More information about the linux-arm-kernel
mailing list