Re: [RFC PATCH v4 11/15] perf stat: Handle taken alone in hardware-grouping

From: Ian Rogers
Date: Tue Mar 26 2024 - 20:05:37 EST

Next message: Kent Overstreet: "[PATCH] fixup! fix missing vmalloc.h includes"
Previous message: Ian Rogers: "Re: [RFC PATCH v4 03/15] perf pmu-events: Add functions in jevent.py to parse counter and event info for hardware aware grouping"
In reply to: Wang, Weilin: "RE: [RFC PATCH v4 11/15] perf stat: Handle taken alone in hardware-grouping"
Next in thread: Wang, Weilin: "RE: [RFC PATCH v4 11/15] perf stat: Handle taken alone in hardware-grouping"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Tue, Mar 26, 2024 at 4:06 PM Wang, Weilin <weilin.wang@xxxxxxxxx> wrote:
>
>
>
> > -----Original Message-----
> > From: Ian Rogers <irogers@xxxxxxxxxx>
> > Sent: Saturday, March 23, 2024 10:25 PM
> > To: Wang, Weilin <weilin.wang@xxxxxxxxx>
> > Cc: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>; Namhyung Kim
> > <namhyung@xxxxxxxxxx>; Arnaldo Carvalho de Melo <acme@xxxxxxxxxx>;
> > Peter Zijlstra <peterz@xxxxxxxxxxxxx>; Ingo Molnar <mingo@xxxxxxxxxx>;
> > Alexander Shishkin <alexander.shishkin@xxxxxxxxxxxxxxx>; Jiri Olsa
> > <jolsa@xxxxxxxxxx>; Hunter, Adrian <adrian.hunter@xxxxxxxxx>; linux-perf-
> > users@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Taylor, Perry
> > <perry.taylor@xxxxxxxxx>; Alt, Samantha <samantha.alt@xxxxxxxxx>; Biggers,
> > Caleb <caleb.biggers@xxxxxxxxx>; Mark Rutland <mark.rutland@xxxxxxx>
> > Subject: Re: [RFC PATCH v4 11/15] perf stat: Handle taken alone in hardware-
> > grouping
> >
> > On Thu, Feb 8, 2024 at 7:14 PM <weilin.wang@xxxxxxxxx> wrote:
> > >
> > > From: Weilin Wang <weilin.wang@xxxxxxxxx>
> > >
> > > Add taken alone into consideration when grouping. Only one taken
> > > alone event is supported per group.
> >
> > Is there an example of this?
>
> Yes, there are about 20+ taken alone core events in SPR. Events like
> UOPS_RETIRED.MS, FRONTEND_RETIRED.DSB_MISS,
> INT_MISC.UNKNOWN_BRANCH_CYCLES, and etc.
>
> If we run `perf stat -e` with two such kind of events, we will see multiplexing.
> ./perf stat -e "INT_MISC.UNKNOWN_BRANCH_CYCLES,UOPS_RETIRED.MS" -a sleep 1

Should these events be in a group?

> Performance counter stats for 'system wide':
>
> 3,257,425 INT_MISC.UNKNOWN_BRANCH_CYCLES (50.04%)
> 134,098,908 UOPS_RETIRED.MS (49.96%)
>
> 1.008902540 seconds time elapsed
>
> This multiplexing should not happen if we run only one such type of event at a time.
> ./perf stat -e "INT_MISC.CLEAR_RESTEER_CYCLES,UOPS_RETIRED.MS" -a sleep 1
>
> Performance counter stats for 'system wide':
>
> 5,107,299 INT_MISC.CLEAR_RESTEER_CYCLES
> 21,929,170 UOPS_RETIRED.MS
>
> 1.009937991 seconds time elapsed
>
> Should I add this example in the commit message?

So taken alone means that the events share a counter? What's going
wrong in the first case and causing the multiplexing?

Thanks,
Ian

> Thanks,
> Weilin
>
> >
> > > Signed-off-by: Weilin Wang <weilin.wang@xxxxxxxxx>
> > > ---
> > > tools/perf/pmu-events/jevents.py | 7 ++++++-
> > > tools/perf/pmu-events/pmu-events.h | 1 +
> > > tools/perf/util/metricgroup.c | 20 +++++++++++++++-----
> > > 3 files changed, 22 insertions(+), 6 deletions(-)
> > >
> > > diff --git a/tools/perf/pmu-events/jevents.py b/tools/perf/pmu-
> > events/jevents.py
> > > index bc91b7efa49a..4fbb367a3228 100755
> > > --- a/tools/perf/pmu-events/jevents.py
> > > +++ b/tools/perf/pmu-events/jevents.py
> > > @@ -56,7 +56,9 @@ _json_event_attributes = [
> > > # The list of counter(s) this event could use
> > > 'counters',
> > > # Longer things (the last won't be iterated over during decompress).
> > > - 'long_desc'
> > > + 'long_desc',
> > > + # Taken alone event could not be collected in the same group with other
> > taken alone event
> > > + 'taken_alone'
> > > ]
> > >
> > > # Attributes that are in pmu_unit_layout.
> > > @@ -355,6 +357,9 @@ class JsonEvent:
> > > self.num_counters = jd.get('NumCounters')
> > > # Number of fixed counter
> > > self.num_fixed_counters = jd.get('NumFixedCounters')
> > > + # If the event is taken alone event, which cannot be grouped with any
> > other
> > > + # taken alone event.
> > > + self.taken_alone = jd.get('TakenAlone')
> > > filter = jd.get('Filter')
> > > self.unit = jd.get('ScaleUnit')
> > > self.perpkg = jd.get('PerPkg')
> > > diff --git a/tools/perf/pmu-events/pmu-events.h b/tools/perf/pmu-
> > events/pmu-events.h
> > > index e245e4738970..837edfeb676a 100644
> > > --- a/tools/perf/pmu-events/pmu-events.h
> > > +++ b/tools/perf/pmu-events/pmu-events.h
> > > @@ -54,6 +54,7 @@ struct pmu_event {
> > > const char *unit;
> > > bool perpkg;
> > > bool deprecated;
> > > + bool taken_alone;
> > > };
> > >
> > > struct pmu_metric {
> > > diff --git a/tools/perf/util/metricgroup.c b/tools/perf/util/metricgroup.c
> > > index fe115f0880f9..95d3868819e3 100644
> > > --- a/tools/perf/util/metricgroup.c
> > > +++ b/tools/perf/util/metricgroup.c
> > > @@ -189,6 +189,7 @@ struct metricgroup__event_info {
> > > /** The event uses special counters that we consider that as free counter
> > > * during the event grouping*/
> > > bool free_counter;
> > > + bool taken_alone;
> > > /** The counters the event allowed to be collected on. */
> > > DECLARE_BITMAP(counters, NR_COUNTERS);
> > > };
> > > @@ -235,6 +236,7 @@ struct metricgroup__group {
> > > DECLARE_BITMAP(fixed_counters, NR_COUNTERS);
> > > /** Head to the list of event names in this group*/
> > > struct list_head event_head;
> > > + bool taken_alone;
> > > };
> > >
> > > struct metricgroup__group_events {
> > > @@ -1717,6 +1719,7 @@ static void metricgroup__free_pmu_info(struct
> > list_head *pmu_info_list)
> > > static struct metricgroup__event_info *event_info__new(const char *name,
> > > const char *pmu_name,
> > > const char *counter,
> > > + bool taken_alone,
> > > bool free_counter)
> > > {
> > > int ret = 0;
> > > @@ -1731,6 +1734,7 @@ static struct metricgroup__event_info
> > *event_info__new(const char *name,
> > > pmu_name = "core";
> > >
> > > e->name = name;
> > > + e->taken_alone = taken_alone;
> > > e->free_counter = free_counter;
> > > e->pmu_name = pmu_name;
> > > if (free_counter) {
> > > @@ -1769,7 +1773,8 @@ static int
> > metricgroup__add_metric_event_callback(const struct pmu_event *pe,
> > > if (!strcasecmp(pe->name, d->event_name)) {
> > > if (!pe->counters)
> > > return -EINVAL;
> > > - event = event_info__new(d->event_id, pe->pmu, pe->counters,
> > /*free_counter=*/false);
> > > + event = event_info__new(d->event_id, pe->pmu, pe->counters,
> > > + pe->taken_alone, /*free_counter=*/false);
> > > if (!event)
> > > return -ENOMEM;
> > > list_add(&event->nd, d->list);
> > > @@ -1892,6 +1897,8 @@ static int find_and_set_counters(struct
> > metricgroup__event_info *e,
> > > int ret;
> > > unsigned long find_bit = 0;
> > >
> > > + if (e->taken_alone && current_group->taken_alone)
> > > + return -ENOSPC;
> > > if (e->free_counter)
> > > return 0;
> > > if (e->fixed_counter) {
> > > @@ -1926,11 +1933,13 @@ static int _insert_event(struct
> > metricgroup__event_info *e,
> > > list_add(&event->nd, &group->event_head);
> > > else
> > > list_add_tail(&event->nd, &group->event_head);
> > > + if (e->taken_alone)
> > > + group->taken_alone = true;
> > > return 0;
> > > }
> > >
> > > /**
> > > - * Insert the new_group node at the end of the group list.
> > > + * Initialize the new group and insert it to the end of the group list.
> > > */
> > > static int insert_new_group(struct list_head *head,
> > > struct metricgroup__group *new_group,
> > > @@ -1940,6 +1949,7 @@ static int insert_new_group(struct list_head
> > *head,
> > > INIT_LIST_HEAD(&new_group->event_head);
> > > fill_counter_bitmap(new_group->gp_counters, 0, num_counters);
> > > fill_counter_bitmap(new_group->fixed_counters, 0,
> > num_fixed_counters);
> > > + new_group->taken_alone = false;
> > > list_add_tail(&new_group->nd, head);
> > > return 0;
> > > }
> > > @@ -2143,8 +2153,8 @@ static int create_grouping(struct list_head
> > *pmu_info_list,
> > > //TODO: for each new core group, we should consider to add events that
> > uses fixed counters
> > > list_for_each_entry(e, event_info_list, nd) {
> > > bitmap_scnprintf(e->counters, NR_COUNTERS, bit_buf,
> > NR_COUNTERS);
> > > - pr_debug("Event name %s, [pmu]=%s, [counters]=%s\n", e->name,
> > > - e->pmu_name, bit_buf);
> > > + pr_debug("Event name %s, [pmu]=%s, [counters]=%s,
> > [taken_alone]=%d\n",
> > > + e->name, e->pmu_name, bit_buf, e->taken_alone);
> > > ret = assign_event_grouping(e, pmu_info_list, &groups);
> > > if (ret)
> > > goto out;
> > > @@ -2191,7 +2201,7 @@ static int hw_aware_build_grouping(struct
> > expr_parse_ctx *ctx __maybe_unused,
> > > if (is_special_event(id)) {
> > > struct metricgroup__event_info *event;
> > >
> > > - event = event_info__new(id, "default_core", "0",
> > > + event = event_info__new(id, "default_core", "0", false,
> >
> > nit: document constant arguments, so "/*taken_alone=*/false,"
> >
> > Thanks,
> > Ian
> >
> > > /*free_counter=*/true);
> > > if (!event)
> > > goto err_out;
> > > --
> > > 2.42.0
> > >

Next message: Kent Overstreet: "[PATCH] fixup! fix missing vmalloc.h includes"
Previous message: Ian Rogers: "Re: [RFC PATCH v4 03/15] perf pmu-events: Add functions in jevent.py to parse counter and event info for hardware aware grouping"
In reply to: Wang, Weilin: "RE: [RFC PATCH v4 11/15] perf stat: Handle taken alone in hardware-grouping"
Next in thread: Wang, Weilin: "RE: [RFC PATCH v4 11/15] perf stat: Handle taken alone in hardware-grouping"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]