Re: [PATCH v8 19/32] x86/resctrl: Complete telemetry event enumeration

[Reinette Chatre]

Hi Tony,

On 8/11/25 11:16 AM, Tony Luck wrote:
> Counters for telemetry events are in MMIO space. Each telemetry_region
> structure returned in the pmt_feature_group returned from OOBMSM contains

"OOBMSM" -> "INTEL_PMT_TELEMETRY"?

> the base MMIO address for the counters.
> 
> There may be multiple aggregators per package. Scan all the
> telemetry_region structures again and save the number of regions together
> with a flex array of the MMIO addresses for each region indexed by
> package id.
> 
> Completed structure for each event group looks like this:
> 
>              +---------------------+---------------------+
> pkginfo** -->|pkginfo[package ID 0]|pkginfo[package ID 1]|
>              +---------------------+---------------------+
>                         |                     |
>                         v                     v
>                 +--------------------+    +--------------------+
>                 |struct pkg_mmio_info|    |struct pkg_mmio_info|
>                 +--------------------+    +--------------------+
>                 |num_regions = M     |    |num_regions = N     |
>                 |  addrs[0]          |    |  addrs[0]          |
>                 |  addrs[1]          |    |  addrs[1]          |
>                 |    ...             |    |    ...             |
>                 |  addrs[M-1]        |    |  addrs[N-1]        |
>                 +--------------------+    +--------------------+
> 
> Build a list (active_event_groups) of all the event groups that
> were successfully enabled. Use it to clean up in intel_aet_exit().

While this adds a note about active_event_groups it does not motivate 
*why* this additional data structure is needed. I find this additional
data structure unnecessary and actually makes the code harder to understand.
As I understand event_group::pfg can already be used to determine if the
event group is active or not. Adding a new data structure to track this thus seems
unnecessary. If this is a data structure created for convenience I think it
should rather be replaced by helpers that use event_group::pfg.

> 
> Signed-off-by: Tony Luck <tony.luck@xxxxxxxxx>
> ---

...
  
> @@ -169,4 +234,13 @@ bool intel_aet_get_events(void)
>  
>  void __exit intel_aet_exit(void)
>  {
> +	struct event_group *evg, *tmp;
> +
> +	list_for_each_entry_safe(evg, tmp, &active_event_groups, list) {
> +		intel_pmt_put_feature_group(evg->pfg);
> +		evg->pfg = NULL;
> +		free_pkg_mmio_info(evg->pkginfo);
> +		evg->pkginfo = NULL;
> +		list_del(&evg->list);
> +	}
>  }

I think above can be simplified by making it symmetrical to
intel_aet_get_events(). 
For example:

static void put_pmt_feature(struct event_group **evgs, unsigned int num_evg)    
{                                                                               
	struct event_group **peg;                                               
                                                                                
	for (peg = evgs; peg < &evgs[num_evg]; peg++) {                         
		if (!(*peg)->pfg)                                               
			return;                                                 
		intel_pmt_put_feature_group((*peg)->pfg);                       
		/* rest of cleanup from intel_aet_exit() above */
	}                                                                       
}                                      

void __ exit intel_aet_exit(void) {                                                          
	put_pmt_feature(known_energy_event_groups,                              
			ARRAY_SIZE(known_energy_event_groups));                 
	put_pmt_feature(known_perf_event_groups,                                
			ARRAY_SIZE(known_perf_event_groups));                   
}                                           

Reinette