lkml.org 
[lkml]   [2022]   [Nov]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH v1 1/9] perf pmu: Add documentation
From


On 2022-11-14 12:04 p.m., Ian Rogers wrote:
> On Mon, Nov 14, 2022 at 7:26 AM Liang, Kan <kan.liang@linux.intel.com> wrote:
>>
>>
>>
>> On 2022-11-14 9:09 a.m., Ian Rogers wrote:
>>> On Mon, Nov 14, 2022 at 5:40 AM Liang, Kan <kan.liang@linux.intel.com> wrote:
>>>>
>>>>
>>>>
>>>> On 2022-11-14 2:51 a.m., Ian Rogers wrote:
>>>>> Add documentation to struct perf_pmu and the associated structs of
>>>>> perf_pmu_alias and perf_pmu_format.
>>>>>
>>>>> Signed-off-by: Ian Rogers <irogers@google.com>
>>>>> ---
>>>>> tools/perf/util/pmu.c | 14 ++++++
>>>>> tools/perf/util/pmu.h | 105 +++++++++++++++++++++++++++++++++++++++---
>>>>> 2 files changed, 113 insertions(+), 6 deletions(-)
>>>>>
>>>>> diff --git a/tools/perf/util/pmu.c b/tools/perf/util/pmu.c
>>>>> index 6a86e6af0903..a8f9f47c6ed9 100644
>>>>> --- a/tools/perf/util/pmu.c
>>>>> +++ b/tools/perf/util/pmu.c
>>>>> @@ -31,10 +31,24 @@
>>>>>
>>>>> struct perf_pmu perf_pmu__fake;
>>>>>
>>>>> +/**
>>>>> + * Values from a format file read from <sysfs>/devices/cpu/format/ held in
>>>>> + * struct perf_pmu. For example, the contents of
>>>>> + * <sysfs>/devices/cpu/format/event may be "config:0-7" and will be represented
>>>>> + * here as name="event", value=PERF_PMU_FORMAT_VALUE_CONFIG and bits 0 to 7 will
>>>>> + * be set.
>>>>> + */
>>>>> struct perf_pmu_format {
>>>>> + /** The modifier/file name. */
>>>>> char *name;
>>>>> + /**
>>>>> + * Which config value the format relates to. Supported values are from
>>>>> + * PERF_PMU_FORMAT_VALUE_CONFIG to PERF_PMU_FORMAT_VALUE_CONFIG_END.
>>>>> + */
>>>>> int value;
>>>>> + /** Which config bits are set by this format value. */
>>>>> DECLARE_BITMAP(bits, PERF_PMU_FORMAT_BITS);
>>>>> + /** Element on list within struct perf_pmu. */
>>>>> struct list_head list;
>>>>> };
>>>>>
>>>>> diff --git a/tools/perf/util/pmu.h b/tools/perf/util/pmu.h
>>>>> index 68e15c38ae71..29571c0f9d15 100644
>>>>> --- a/tools/perf/util/pmu.h
>>>>> +++ b/tools/perf/util/pmu.h
>>>>> @@ -34,30 +34,91 @@ struct perf_pmu_caps {
>>>>> };
>>>>>
>>>>> struct perf_pmu {
>>>>> + /** The name of the PMU such as "cpu". */
>>>>> char *name;
>>>>> + /**
>>>>> + * Optional alternate name for the PMU determined in architecture
>>>>> + * specific code.
>>>>> + */
>>>>> char *alias_name;
>>>>> + /**
>>>>> + * Optional PMU identifier read from
>>>>> + * <sysfs>/bus/event_source/devices/<name>/identifier.
>>>>> + */
>>>>> char *id;
>>>>> + /**
>>>>> + * Perf event attributed type value, read from
>>>>> + * <sysfs>/bus/event_source/devices/<name>/type.
>>>>> + */
>>>>> __u32 type;
>>>>> + /**
>>>>> + * Can the PMU name be selected as if it were an event?
>>>>> + */
>>>>> bool selectable;
>>>>> + /**
>>>>> + * Is the PMU not within the CPU core? Determined by the presence of
>>>>> + * <sysfs>/bus/event_source/devices/<name>/cpumask.
>>>>> + */
>>>>> bool is_uncore;
>>>>> + /** Is the PMU name either cpu_core or cpu_atom. */
>>>>
>>>> I don't think we want to limit the hybrid names only to cpu_core or
>>>> cpu_atom. Maybe something as below?
>>>> /* Is a hybrid CPU PMU, e.g., cpu_core, cpu_atom. */
>>>
>>> Currently the hybrid code only works for cpu_core or cpu_atom, a
>>> limitation of its implementation.
>>
>> I don't think so. See perf_pmu__hybrid_mounted(). If a PMU is named as
>> "cpu_", we treat it as a hybrid CPU PMU. The cpu_core or cpu_atom should
>> only be hard coded to specially handle some model-specific cases, e.g.,
>> mem-loads-aux event.
>
> Ugh.. Why is a property about something being a CPU named in an Intel
> specific term for big little? Why wasn't this generalized? Why are CPU
> PMUs assumed to be prefixed cpu_* .. ? The comment is factually
> correct, we can hold off deprecating it on the assumption that it'd be
> renamed to something more appropriate like is_cpu.

Does any code really use the pmu->is_hybrid?
From what I read, it's only used in the pmu_lookup(), where it's
assigned. It looks like we don't need it in the struct perf_pmu. If so,
let's remove it completely from struct perf_pmu.

Thanks,
Kan
>
> Thanks,
> Ian
>
>> Thanks,
>> Kan
>>
>>> As pointed out in a later patch,
>>> this bool isn't being used when it could be and I think we can work to
>>> remove it. It would be possible to remove all uses of this with
>>> perf_pmu__is_hybrid. As such I think it may be useful to mark the
>>> hybrid variables in struct perf_pmu as deprecated while we work to
>>> replace their use with more generic just any PMU code.
>>>
>>> Thanks,
>>> Ian
>>>
>>>> Thanks,
>>>> Kan
>>>>
>>>>> bool is_hybrid;
>>>>> + /**
>>>>> + * Are events auxiliary events? Determined in architecture specific
>>>>> + * code.
>>>>> + */
>>>>> bool auxtrace;
>>>>> + /**
>>>>> + * Number of levels of :ppp precision supported by the PMU, read from
>>>>> + * <sysfs>/bus/event_source/devices/<name>/caps/max_precise.
>>>>> + */
>>>>> int max_precise;
>>>>> + /**
>>>>> + * Optional default perf_event_attr determined in architecture specific
>>>>> + * code.
>>>>> + */
>>>>> struct perf_event_attr *default_config;
>>>>> + /**
>>>>> + * Empty or the contents of either of:
>>>>> + * <sysfs>/bus/event_source/devices/<name>/cpumask.
>>>>> + * <sysfs>/bus/event_source/devices/<cpu>/cpus.
>>>>> + */
>>>>> struct perf_cpu_map *cpus;
>>>>> - struct list_head format; /* HEAD struct perf_pmu_format -> list */
>>>>> - struct list_head aliases; /* HEAD struct perf_pmu_alias -> list */
>>>>> + /**
>>>>> + * Holds the contents of files read from
>>>>> + * <sysfs>/bus/event_source/devices/<name>/format/. The contents specify
>>>>> + * which event parameter changes what config, config1 or config2 bits.
>>>>> + */
>>>>> + struct list_head format;
>>>>> + /**
>>>>> + * List of struct perf_pmu_alias. Each alias corresponds to an event
>>>>> + * read from <sysfs>/bus/event_source/devices/<name>/events/ or from
>>>>> + * json events in pmu-events.c.
>>>>> + */
>>>>> + struct list_head aliases;
>>>>> + /** Has the list caps been initialized? */
>>>>> bool caps_initialized;
>>>>> + /** The length of the list caps. */
>>>>> u32 nr_caps;
>>>>> - struct list_head caps; /* HEAD struct perf_pmu_caps -> list */
>>>>> - struct list_head list; /* ELEM */
>>>>> + /**
>>>>> + * Holds the contents of files read from
>>>>> + * <sysfs>/bus/event_source/devices/<name>/caps/. The contents are pairs
>>>>> + * of the filename with the value of its contents, for example,
>>>>> + * max_precise (see above) may have a value of 3.
>>>>> + */
>>>>> + struct list_head caps;
>>>>> + /** Element on pmus list in pmu.c. */
>>>>> + struct list_head list;
>>>>> + /** Element on perf_pmu__hybrid_pmus. */
>>>>> struct list_head hybrid_list;
>>>>>
>>>>> + /** Features to inhibit when events on this PMU are opened. */
>>>>> struct {
>>>>> + /** Disables perf_event_attr exclude_guest and exclude_host. */
>>>>> bool exclude_guest;
>>>>> } missing_features;
>>>>> };
>>>>>
>>>>> +/** A special global PMU used for testing. */
>>>>> extern struct perf_pmu perf_pmu__fake;
>>>>>
>>>>> struct perf_pmu_info {
>>>>> @@ -71,21 +132,53 @@ struct perf_pmu_info {
>>>>>
>>>>> #define UNIT_MAX_LEN 31 /* max length for event unit name */
>>>>>
>>>>> +/**
>>>>> + * An event either read from sysfs or builtin in pmu-events.c, created by
>>>>> + * parsing the pmu-events json files.
>>>>> + */
>>>>> struct perf_pmu_alias {
>>>>> char *name;
>>>>> + /** Optional short description of the event. */
>>>>> char *desc;
>>>>> + /** Optional long description. */
>>>>> char *long_desc;
>>>>> + /**
>>>>> + * Optional topic such as cache or pipeline, particularly for json
>>>>> + * events.
>>>>> + */
>>>>> char *topic;
>>>>> + /** Comma separated parameter list. */
>>>>> char *str;
>>>>> - struct list_head terms; /* HEAD struct parse_events_term -> list */
>>>>> - struct list_head list; /* ELEM */
>>>>> + /** Owned list of the original parsed parameters. */
>>>>> + struct list_head terms;
>>>>> + /** List element of struct perf_pmu aliases. */
>>>>> + struct list_head list;
>>>>> + /** Units for the event, such as bytes or cache lines. */
>>>>> char unit[UNIT_MAX_LEN+1];
>>>>> + /** Value to scale read counter values by. */
>>>>> double scale;
>>>>> + /**
>>>>> + * Does the file
>>>>> + * <sysfs>/bus/event_source/devices/<pmu_name>/events/<name>.per-pkg or
>>>>> + * equivalent json value exist and have the value 1.
>>>>> + */
>>>>> bool per_pkg;
>>>>> + /**
>>>>> + * Does the file
>>>>> + * <sysfs>/bus/event_source/devices/<pmu_name>/events/<name>.snapshot
>>>>> + * exist and have the value 1.
>>>>> + */
>>>>> bool snapshot;
>>>>> + /** Is the event hidden and so not shown in perf list by default. */
>>>>> bool deprecated;
>>>>> + /**
>>>>> + * A metric expression associated with an event. Doing this makes little
>>>>> + * sense due to scale and unit applying to both.
>>>>> + */
>>>>> char *metric_expr;
>>>>> + /** A name for the metric. unit applying to both. */
>>>>> char *metric_name;
>>>>> + /** The name copied from struct perf_pmu. */
>>>>> char *pmu_name;
>>>>> };
>>>>>

\
 
 \ /
  Last update: 2022-11-14 19:51    [W:2.337 / U:0.004 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site