Re: [PATCH v1 1/2] perf/core: Add API to look up PMU type by name

[Peter Zijlstra]

On Fri, Feb 23, 2018 at 04:19:37PM -0800, Saravana Kannan wrote:
> When the event numbers registered by multiple PMUs overlap, the
> attr->type value passed to perf_event_create_kernel_counter() is used
> to determine which PMU to use to create a perf_event.
> 
> However, when the PMU in question is not a standard PMU (defined in
> perf_type_id), there is no way for a kernel client to look up the PMU
> type for the PMU of interest and set the attr->type appropriately.
> 
> So, add an API to look up the PMU type by name. That way, the kernel
> APIs can function in a fashion similar to the user space interface.
> 
> Signed-off-by: Saravana Kannan <skannan@xxxxxxxxxxxxxx>
> ---
>  kernel/events/core.c | 23 +++++++++++++++++++++++
>  1 file changed, 23 insertions(+)
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 96db9ae..5d3df58 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -10310,6 +10310,29 @@ static int perf_event_set_clock(struct perf_event *event, clockid_t clk_id)
>  	return err;
>  }
>  
> +int perf_find_pmu_type_by_name(const char *name)
> +{
> +	struct pmu *pmu;
> +	int ret = -1;
> +
> +	mutex_lock(&pmus_lock);
> +
> +	list_for_each_entry(pmu, &pmus, entry) {
> +		if (!pmu->name || pmu->type < 0)
> +			continue;
> +
> +		if (!strcmp(name, pmu->name)) {
> +			ret = pmu->type;
> +			goto out;
> +		}
> +	}
> +
> +out:
> +	mutex_unlock(&pmus_lock);
> +
> +	return ret;
> +}

Not without an in-tree user.