Re: [PATCH V2 5/5] perf mem: Clean up is_mem_loads_aux_event()

[Liang, Kan]

On 2023-12-09 1:27 a.m., Leo Yan wrote:
> On Thu, Dec 07, 2023 at 11:23:38AM -0800, kan.liang@xxxxxxxxxxxxxxx wrote:
>> From: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
>>
>> The aux_event can be retrieved from the perf_pmu now. Implement a
>> generic support.
>>
>> Reviewed-by: Ian Rogers <irogers@xxxxxxxxxx>
>> Tested-by: Ravi Bangoria <ravi.bangoria@xxxxxxx>
>> Signed-off-by: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
>> ---
>>  tools/perf/arch/x86/util/mem-events.c | 23 ++++-------------------
>>  tools/perf/util/mem-events.c          | 14 ++++++++++++--
>>  2 files changed, 16 insertions(+), 21 deletions(-)
>>
>> diff --git a/tools/perf/arch/x86/util/mem-events.c b/tools/perf/arch/x86/util/mem-events.c
>> index b776d849fc64..62df03e91c7e 100644
>> --- a/tools/perf/arch/x86/util/mem-events.c
>> +++ b/tools/perf/arch/x86/util/mem-events.c
>> @@ -1,11 +1,9 @@
>>  // SPDX-License-Identifier: GPL-2.0
>> -#include "util/pmu.h"
>> -#include "util/pmus.h"
>> -#include "util/env.h"
>> -#include "map_symbol.h"
>> -#include "mem-events.h"
>>  #include "linux/string.h"
>> -#include "env.h"
>> +#include "util/map_symbol.h"
>> +#include "util/mem-events.h"
>> +#include "mem-events.h"
>> +
>>  
>>  #define MEM_LOADS_AUX		0x8203
>>  
>> @@ -28,16 +26,3 @@ struct perf_mem_event perf_mem_events_amd[PERF_MEM_EVENTS__MAX] = {
>>  	E(NULL,		NULL,		NULL,	false,	0),
>>  	E("mem-ldst",	"%s//",		NULL,	false,	0),
>>  };
>> -
>> -bool is_mem_loads_aux_event(struct evsel *leader)
>> -{
>> -	struct perf_pmu *pmu = perf_pmus__find("cpu");
>> -
>> -	if (!pmu)
>> -		pmu = perf_pmus__find("cpu_core");
>> -
>> -	if (pmu && !perf_pmu__have_event(pmu, "mem-loads-aux"))
>> -		return false;
>> -
>> -	return leader->core.attr.config == MEM_LOADS_AUX;
>> -}
>> diff --git a/tools/perf/util/mem-events.c b/tools/perf/util/mem-events.c
>> index 0d174f161034..d418320e52e3 100644
>> --- a/tools/perf/util/mem-events.c
>> +++ b/tools/perf/util/mem-events.c
>> @@ -103,9 +103,19 @@ static const char *perf_pmu__mem_events_name(int i, struct perf_pmu *pmu)
>>  	return NULL;
>>  }
>>  
>> -__weak bool is_mem_loads_aux_event(struct evsel *leader __maybe_unused)
>> +bool is_mem_loads_aux_event(struct evsel *leader)
>>  {
>> -	return false;
>> +	struct perf_pmu *pmu = leader->pmu;
>> +	struct perf_mem_event *e;
>> +
>> +	if (!pmu || !pmu->mem_events)
>> +		return false;
>> +
>> +	e = &pmu->mem_events[PERF_MEM_EVENTS__LOAD];
>> +	if (!e->aux_event)
>> +		return false;
>> +
>> +	return leader->core.attr.config == e->aux_event;
>>  }
> 
> I am wandering if we need to set the field 'aux_event' for Arm SPE.
> 
> So a quesiton maybe is not relevant with this patch actually, we can
> see is_mem_loads_aux_event() is invoked in the file util/record.c:
> 
>   static struct evsel *evsel__read_sampler(struct evsel *evsel, struct evlist *evlist)
>   {
>           struct evsel *leader = evsel__leader(evsel);
> 
>           if (evsel__is_aux_event(leader) || arch_topdown_sample_read(leader) ||
>               is_mem_loads_aux_event(leader)) {
>               ...
>           }
>   
>           return leader;
>   }
> 
> Has evsel__is_aux_event() covered the memory load aux event?  If it's,
> then is_mem_loads_aux_event() is not needed anymore.

They are two different things. The evsel__is_aux_event() should means an
event requires AUX area, like intel_pt.

While the aux event for the mem_loads event is an extra event which has
to be scheduled together with the mem_loads event when sampling. It's
only available for some Intel platforms, e.g., SPR.

Thanks,
Kan
> 
> Thanks,
> Leo