lkml.org 
[lkml]   [2021]   [Aug]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
SubjectRe: [PATCH v4 1/2] perf pmu: Add PMU alias support
From
Date
Hi John,

On 8/12/2021 8:14 PM, John Garry wrote:
> perf pmu: Add x86 PMU alias support
>
>> +char *pmu_find_real_name(const char *name)
>> +{
>> +    static bool cached_list;
>> +
>> +    if (cached_list)
>> +        return __pmu_find_real_name(name);
>> +
>> +    setup_pmu_alias_list();
>> +    cached_list = true;
>> +
>> +    return __pmu_find_real_name(name);
>> +}
>> +
>> +char *pmu_find_alias_name(const char *name)
>> +{
>> +    struct perf_pmu_alias_name *pmu;
>> +
>> +    list_for_each_entry(pmu, &pmu_alias_name_list, list) {
>> +        if (!strcmp(name, pmu->name))
>> +            return strdup(pmu->alias);
>
> I would not expect a function which does a "find" to duplicate the name.
>
> Same goes for all the other places which does similar.
>

Yes, you are right. Jiri has the similar comment for the name duplicating here. We will still keep
the 'pmu->name = strdup(name);' in pmu_lookup() in v5.

>> +    }
>> +    return NULL;
>> +}
>> diff --git a/tools/perf/util/parse-events.y b/tools/perf/util/parse-events.y
>> index 9321bd0e2f76..d94e48e1ff9b 100644
>> --- a/tools/perf/util/parse-events.y
>> +++ b/tools/perf/util/parse-events.y
>> @@ -316,7 +316,8 @@ event_pmu_name opt_pmu_config
>>               if (!strncmp(name, "uncore_", 7) &&
>>                   strncmp($1, "uncore_", 7))
>>                   name += 7;
>> -            if (!perf_pmu__match(pattern, name, $1)) {
>> +            if (!perf_pmu__match(pattern, name, $1) ||
>> +                !perf_pmu__match(pattern, pmu->alias_name, $1)) {
>>                   if (parse_events_copy_term_list(orig_terms, &terms))
>>                       CLEANUP_YYABORT;
>>                   if (!parse_events_add_pmu(_parse_state, list, pmu->name, terms, true, false))
>> diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
>> index fc683bc41715..796a4be752f4 100644
>> --- a/tools/perf/util/pmu.c
>> +++ b/tools/perf/util/pmu.c
>> @@ -946,6 +946,18 @@ perf_pmu__get_default_config(struct perf_pmu *pmu __maybe_unused)
>>       return NULL;
>>   }
>> +char * __weak
>> +pmu_find_real_name(const char *name)
>> +{
>> +    return strdup(name);
>> +}
>
> That's not finding anything.
>

Will just return name in non-x86 implementation.

pmu_find_real_name(const char *name)
{
return name;
}

>> +
>> +char * __weak
>> +pmu_find_alias_name(const char *name __maybe_unused)
>> +{
>> +    return NULL;
> > +}
>> +
>>   static int pmu_max_precise(const char *name)
>>   {
>>       char path[PATH_MAX];
>> @@ -959,19 +971,25 @@ static int pmu_max_precise(const char *name)
>>       return max_precise;
>>   }
>> -static struct perf_pmu *pmu_lookup(const char *name)
>> +static struct perf_pmu *pmu_lookup(const char *lookup_name)
>>   {
>> -    struct perf_pmu *pmu;
>> +    struct perf_pmu *pmu = NULL;
>>       LIST_HEAD(format);
>>       LIST_HEAD(aliases);
>>       __u32 type;
>> -    bool is_hybrid = perf_pmu__hybrid_mounted(name);
>> +    bool is_hybrid;
>> +    char *name = pmu_find_real_name(lookup_name);
>> +
>> +    if (!name)
>> +        return NULL;
>> +
>> +    is_hybrid = perf_pmu__hybrid_mounted(name);
>>       /*
>>        * Check pmu name for hybrid and the pmu may be invalid in sysfs
>>        */
>>       if (!strncmp(name, "cpu_", 4) && !is_hybrid)
>> -        return NULL;
>> +        goto out;
>>       /*
>>        * The pmu data we store & need consists of the pmu
>> @@ -979,23 +997,24 @@ static struct perf_pmu *pmu_lookup(const char *name)
>>        * now.
>>        */
>>       if (pmu_format(name, &format))
>> -        return NULL;
>> +        goto out;
>>       /*
>>        * Check the type first to avoid unnecessary work.
>>        */
>>       if (pmu_type(name, &type))
>> -        return NULL;
>> +        goto out;
>>       if (pmu_aliases(name, &aliases))
>> -        return NULL;
>> +        goto out;
>>       pmu = zalloc(sizeof(*pmu));
>>       if (!pmu)
>> -        return NULL;
>> +        goto out;
>>       pmu->cpus = pmu_cpumask(name);
>> -    pmu->name = strdup(name);
>> +    pmu->name = name;
>> +    pmu->alias_name = pmu_find_alias_name(name);
>>       pmu->type = type;
>>       pmu->is_uncore = pmu_is_uncore(name);
>>       if (pmu->is_uncore)
>> @@ -1017,6 +1036,10 @@ static struct perf_pmu *pmu_lookup(const char *name)
>>       pmu->default_config = perf_pmu__get_default_config(pmu);
>> +out:
>> +    if (!pmu)
>> +        free(name);
>
> I don't understand this. There are lots of places this function can fail , but we don't free
> memories allocated previously - why just free this one?
>

That's because it's only introduced by my patch (duplicating name). :)

But if we keep 'pmu->name = strdup(name)' in pmu_lookup(), I'm OK to keep original code/behavior.

>> +
>>       return pmu;
>>   }
>> @@ -1025,7 +1048,8 @@ static struct perf_pmu *pmu_find(const char *name)
>>       struct perf_pmu *pmu;
>>       list_for_each_entry(pmu, &pmus, list)
>> -        if (!strcmp(pmu->name, name))
>> +        if (!strcmp(pmu->name, name) ||
>> +            (pmu->alias_name && !strcmp(pmu->alias_name, name)))
>>               return pmu;
>
> I'd be inclined to use {} for the list_for_each_entry() call
>

OK, I will add {}, thanks!

>>       return NULL;
>> @@ -1920,6 +1944,9 @@ bool perf_pmu__has_hybrid(void)
>>   int perf_pmu__match(char *pattern, char *name, char *tok)
>>   {
>> +    if (!name)
>> +        return -1;
>> +
>>       if (fnmatch(pattern, name, 0))
>>           return -1;
>> diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
>> index 926da483a141..f6ca9f6a06ef 100644
>> --- a/tools/perf/util/pmu.h
>> +++ b/tools/perf/util/pmu.h
>> @@ -21,6 +21,7 @@ enum {
>>   #define PERF_PMU_FORMAT_BITS 64
>>   #define EVENT_SOURCE_DEVICE_PATH "/bus/event_source/devices/"
>>   #define CPUS_TEMPLATE_CPU    "%s/bus/event_source/devices/%s/cpus"
>> +#define MAX_PMU_NAME_LEN 128
>>   struct perf_event_attr;
>> @@ -32,6 +33,7 @@ struct perf_pmu_caps {
>>   struct perf_pmu {
>>       char *name;
>> +    char *alias_name;    /* PMU alias name */
>
> useless comment
>

OK, remove it :)

Thanks
Jin Yao

>>       char *id;
>>       __u32 type;
>>       bool selectable;
>> @@ -135,4 +137,7 @@ void perf_pmu__warn_invalid_config(struct perf_pmu *pmu, __u64 config,
>>   bool perf_pmu__has_hybrid(void);
>>   int perf_pmu__match(char *pattern, char *name, char *tok);
>> +char *pmu_find_real_name(const char *name);
>> +char *pmu_find_alias_name(const char *name);
>> +
>>   #endif /* __PMU_H */
>>
>

\
 
 \ /
  Last update: 2021-08-13 03:59    [W:0.049 / U:0.520 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site