Re: [PATCH v4 2/6] perf: imx_perf: refactor driver for imx93

[Will Deacon]

On Wed, Jan 31, 2024 at 01:58:07PM +0800, Xu Yang wrote:
> This driver is initinally used to support imx93 Soc and now it's time to
> add support for imx95 Soc. However, some macro definitions and events are
> different on these two Socs. For preparing imx95 supports, this will
> refactor driver for imx93.
> 
> Signed-off-by: Xu Yang <xu.yang_2@xxxxxxx>
> 
> ---
> Changes in v4:
>  - new patch
> ---
>  drivers/perf/fsl_imx9_ddr_perf.c | 121 ++++++++++++++++++++++---------
>  1 file changed, 87 insertions(+), 34 deletions(-)

[...]

> @@ -476,12 +490,12 @@ static int ddr_perf_event_add(struct perf_event *event, int flags)
>  	hwc->idx = counter;
>  	hwc->state |= PERF_HES_STOPPED;
>  
> +	/* read trans, write trans, read beat */
> +	imx93_ddr_perf_monitor_config(pmu, cfg, cfg1, cfg2);
> +
>  	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);
> -
>  	return 0;

This change looks like more than just refactoring and should probably be a
separate patch. Is it a bug fix for the existing code?

> +static int ddr_perf_add_events(struct ddr_pmu *pmu)
> +{
> +	int i, ret;
> +	struct attribute **attrs = pmu->devtype_data->attrs;
> +	struct device *pmu_dev = pmu->pmu.dev;
> +
> +	if (!attrs)
> +		return 0;
> +
> +	for (i = 0; attrs[i]; 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;

Can you use the '.is_visible' callback in 'struct attribute_group' to avoid
having to manipulate sysfs directly like this? For example, create separate
groups for the imx93 and imx95-specific events and only make them visible
if we're on the appropriate hardware.

Will