Re: [PATCH v2 12/32] perf/x86/intel/cqm: schedule work for rotation task

[Thomas Gleixner]

On Wed, 11 May 2016, David Carrillo-Cisneros wrote:

> Make execution of rotation a delayed_work that does a best effort
> to rotate __cqm_min_progress_rate pmonrs per-second in every package.
> 
> Reviewed-by: Stephane Eranian <eranian@xxxxxxxxxx>
> Signed-off-by: David Carrillo-Cisneros <davidcc@xxxxxxxxxx>
> ---
>  arch/x86/events/intel/cqm.c | 70 ++++++++++++++++++++++++++++++++++++++++++++-
>  arch/x86/events/intel/cqm.h | 21 ++++++++++++++
>  2 files changed, 90 insertions(+), 1 deletion(-)
> 
> diff --git a/arch/x86/events/intel/cqm.c b/arch/x86/events/intel/cqm.c
> index 203fc66..a61dd70 100644
> --- a/arch/x86/events/intel/cqm.c
> +++ b/arch/x86/events/intel/cqm.c
> @@ -222,6 +222,8 @@ static int pkg_data_init_cpu(int cpu)
>  	mutex_init(&pkg_data->pkg_data_mutex);
>  	raw_spin_lock_init(&pkg_data->pkg_data_lock);
>  
> +	INIT_DELAYED_WORK(
> +		&pkg_data->rotation_work, intel_cqm_rmid_rotation_work);

How's that supposed to compile? intel_cqm_rmid_rotation_work is defined below.

> +/*
> + * Schedule rotation and rmid's timed update in all packages.
> + * Reescheduling will stop when no longer needed.
> + */
> +static void intel_cqm_schedule_work_all_pkgs(void)
> +{
> +	int pkg_id;
> +
> +	cqm_pkg_id_for_each_online(pkg_id)
> +		__intel_cqm_schedule_rotation_for_pkg(pkg_id);

That's going to make the NOHZ_FULL people very happy. NOT!

> @@ -1744,6 +1810,8 @@ static int intel_cqm_event_init(struct perf_event *event)
>  
>  	mutex_unlock(&cqm_mutex);
>  
> +	intel_cqm_schedule_work_all_pkgs();

Why would you trigger rotation on each package if you just monitor a given
package?

Thanks,

	tglx