Re: [PATCH 2/7] perf: Add PERF_EV_CAP_READ_SCOPE

[Peter Zijlstra <peterz@infradead.org>]

On Fri, Aug 02, 2024 at 08:16:38AM -0700, kan.liang@linux.intel.com wrote:
> From: Kan Liang <kan.liang@linux.intel.com>
> 
> Usually, an event can be read from any CPU of the scope. It doesn't need
> to be read from the advertised CPU.
> 
> Add a new event cap, PERF_EV_CAP_READ_SCOPE. An event of a PMU with
> scope can be read from any active CPU in the scope.
> 
> Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
> ---
>  include/linux/perf_event.h |  3 +++
>  kernel/events/core.c       | 14 +++++++++++---
>  2 files changed, 14 insertions(+), 3 deletions(-)
> 
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index 1102d5c2be70..1206bc86eb4f 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -633,10 +633,13 @@ typedef void (*perf_overflow_handler_t)(struct perf_event *,
>   * PERF_EV_CAP_SIBLING: An event with this flag must be a group sibling and
>   * cannot be a group leader. If an event with this flag is detached from the
>   * group it is scheduled out and moved into an unrecoverable ERROR state.
> + * PERF_EV_CAP_READ_SCOPE: A CPU event that can be read from any CPU of the
> + * PMU scope where it is active.
>   */
>  #define PERF_EV_CAP_SOFTWARE		BIT(0)
>  #define PERF_EV_CAP_READ_ACTIVE_PKG	BIT(1)
>  #define PERF_EV_CAP_SIBLING		BIT(2)
> +#define PERF_EV_CAP_READ_SCOPE		BIT(3)
>  
>  #define SWEVENT_HLIST_BITS		8
>  #define SWEVENT_HLIST_SIZE		(1 << SWEVENT_HLIST_BITS)
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 5e1877c4cb4c..c55294f34575 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -4463,16 +4463,24 @@ struct perf_read_data {
>  	int ret;
>  };
>  
> +static inline const struct cpumask *perf_scope_cpu_topology_cpumask(unsigned int scope, int cpu);
> +
>  static int __perf_event_read_cpu(struct perf_event *event, int event_cpu)
>  {
> +	int local_cpu = smp_processor_id();
>  	u16 local_pkg, event_pkg;
>  
>  	if ((unsigned)event_cpu >= nr_cpu_ids)
>  		return event_cpu;
>  
> -	if (event->group_caps & PERF_EV_CAP_READ_ACTIVE_PKG) {
> -		int local_cpu = smp_processor_id();
> +	if (event->group_caps & PERF_EV_CAP_READ_SCOPE) {
> +		const struct cpumask *cpumask = perf_scope_cpu_topology_cpumask(event->pmu->scope, event_cpu);
> +
> +		if (cpumask && cpumask_test_cpu(local_cpu, cpumask))
> +			return local_cpu;
> +	}
>  
> +	if (event->group_caps & PERF_EV_CAP_READ_ACTIVE_PKG) {

I'm guessing the goal is to eventually remove this one, right?