[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAP-5=fUVu1nhSZ2KehLdEMTHCW393EXuk3UjnNjwBpe4EOMt4w@mail.gmail.com>
Date: Mon, 7 Apr 2025 13:10:02 -0700
From: Ian Rogers <irogers@...gle.com>
To: "Liang, Kan" <kan.liang@...ux.intel.com>
Cc: Peter Zijlstra <peterz@...radead.org>, Ingo Molnar <mingo@...hat.com>,
Arnaldo Carvalho de Melo <acme@...nel.org>, Namhyung Kim <namhyung@...nel.org>,
Mark Rutland <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>, Jiri Olsa <jolsa@...nel.org>,
Adrian Hunter <adrian.hunter@...el.com>, Weilin Wang <weilin.wang@...el.com>,
James Clark <james.clark@...aro.org>, Xu Yang <xu.yang_2@....com>,
John Garry <john.g.garry@...cle.com>, Howard Chu <howardchu95@...il.com>,
Levi Yun <yeoreum.yun@....com>, Dominique Martinet <asmadeus@...ewreck.org>,
linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2 11/16] perf intel-tpebs: Add mutex for tpebs_results
On Mon, Apr 7, 2025 at 12:49 PM Liang, Kan <kan.liang@...ux.intel.com> wrote:
>
>
>
> On 2025-04-07 1:00 a.m., Ian Rogers wrote:
> > Ensure sample reader isn't racing with events being added/removed.
>
> Is it a bug fix?
>
> I think the tpebs have been merged for a while. Should the backport be
> required?
Because things like interval mode aren't currently working with tpebs
I think the issues that could arise are minimal. A fixes tag would
drag in this whole series and I'm not sure that's what we want.
Thanks,
Ian
> Thanks,
> Kan>
> > Signed-off-by: Ian Rogers <irogers@...gle.com>
> > ---
> > tools/perf/util/intel-tpebs.c | 51 ++++++++++++++++++++++++++++++-----
> > 1 file changed, 44 insertions(+), 7 deletions(-)
> >
> > diff --git a/tools/perf/util/intel-tpebs.c b/tools/perf/util/intel-tpebs.c
> > index 452ce3698221..29616d087bce 100644
> > --- a/tools/perf/util/intel-tpebs.c
> > +++ b/tools/perf/util/intel-tpebs.c
> > @@ -16,6 +16,7 @@
> > #include "debug.h"
> > #include "evlist.h"
> > #include "evsel.h"
> > +#include "mutex.h"
> > #include "session.h"
> > #include "tool.h"
> > #include "cpumap.h"
> > @@ -32,6 +33,7 @@ bool tpebs_recording;
> > static LIST_HEAD(tpebs_results);
> > static pthread_t tpebs_reader_thread;
> > static struct child_process tpebs_cmd;
> > +static struct mutex tpebs_mtx;
> >
> > struct tpebs_retire_lat {
> > struct list_head nd;
> > @@ -51,6 +53,19 @@ struct tpebs_retire_lat {
> >
> > static struct tpebs_retire_lat *tpebs_retire_lat__find(struct evsel *evsel);
> >
> > +static void tpebs_mtx_init(void)
> > +{
> > + mutex_init(&tpebs_mtx);
> > +}
> > +
> > +static struct mutex *tpebs_mtx_get(void)
> > +{
> > + static pthread_once_t tpebs_mtx_once = PTHREAD_ONCE_INIT;
> > +
> > + pthread_once(&tpebs_mtx_once, tpebs_mtx_init);
> > + return &tpebs_mtx;
> > +}
> > +
> > static int evsel__tpebs_start_perf_record(struct evsel *evsel, int control_fd[], int ack_fd[])
> > {
> > const char **record_argv;
> > @@ -59,13 +74,15 @@ static int evsel__tpebs_start_perf_record(struct evsel *evsel, int control_fd[],
> > char cpumap_buf[50];
> > struct tpebs_retire_lat *t;
> >
> > + mutex_lock(tpebs_mtx_get());
> > list_for_each_entry(t, &tpebs_results, nd)
> > tpebs_event_size++;
> >
> > record_argv = malloc((10 + 2 * tpebs_event_size) * sizeof(*record_argv));
> > - if (!record_argv)
> > + if (!record_argv) {
> > + mutex_unlock(tpebs_mtx_get());
> > return -ENOMEM;
> > -
> > + }
> > record_argv[i++] = "perf";
> > record_argv[i++] = "record";
> > record_argv[i++] = "-W";
> > @@ -101,6 +118,7 @@ static int evsel__tpebs_start_perf_record(struct evsel *evsel, int control_fd[],
> > list_for_each_entry(t, &tpebs_results, nd)
> > t->started = true;
> >
> > + mutex_unlock(tpebs_mtx_get());
> > return ret;
> > }
> >
> > @@ -112,9 +130,12 @@ static int process_sample_event(const struct perf_tool *tool __maybe_unused,
> > {
> > struct tpebs_retire_lat *t;
> >
> > + mutex_lock(tpebs_mtx_get());
> > t = tpebs_retire_lat__find(evsel);
> > - if (!t)
> > + if (!t) {
> > + mutex_unlock(tpebs_mtx_get());
> > return -EINVAL;
> > + }
> > /*
> > * Need to handle per core results? We are assuming average retire
> > * latency value will be used. Save the number of samples and the sum of
> > @@ -123,6 +144,7 @@ static int process_sample_event(const struct perf_tool *tool __maybe_unused,
> > t->count += 1;
> > t->sum += sample->retire_lat;
> > t->val = (double) t->sum / t->count;
> > + mutex_unlock(tpebs_mtx_get());
> > return 0;
> > }
> >
> > @@ -229,7 +251,6 @@ static struct tpebs_retire_lat *tpebs_retire_lat__new(struct evsel *evsel)
> > return NULL;
> > }
> > result->evsel = evsel;
> > - list_add_tail(&result->nd, &tpebs_results);
> > return result;
> > }
> >
> > @@ -282,16 +303,22 @@ static struct tpebs_retire_lat *tpebs_retire_lat__find(struct evsel *evsel)
> > static int evsel__tpebs_prepare(struct evsel *evsel)
> > {
> > struct evsel *pos;
> > - struct tpebs_retire_lat *tpebs_event = tpebs_retire_lat__find(evsel);
> > + struct tpebs_retire_lat *tpebs_event;
> >
> > + mutex_lock(tpebs_mtx_get());
> > + tpebs_event = tpebs_retire_lat__find(evsel);
> > if (tpebs_event) {
> > /* evsel, or an identically named one, was already prepared. */
> > + mutex_unlock(tpebs_mtx_get());
> > return 0;
> > }
> > tpebs_event = tpebs_retire_lat__new(evsel);
> > if (!tpebs_event)
> > return -ENOMEM;
> >
> > + list_add_tail(&tpebs_event->nd, &tpebs_results);
> > + mutex_unlock(tpebs_mtx_get());
> > +
> > /*
> > * Eagerly prepare all other evsels on the list to try to ensure that by
> > * open they are all known.
> > @@ -317,6 +344,7 @@ static int evsel__tpebs_prepare(struct evsel *evsel)
> > int evsel__tpebs_open(struct evsel *evsel)
> > {
> > int ret;
> > + bool tpebs_empty;
> >
> > /* We should only run tpebs_start when tpebs_recording is enabled. */
> > if (!tpebs_recording)
> > @@ -336,7 +364,10 @@ int evsel__tpebs_open(struct evsel *evsel)
> > if (ret)
> > return ret;
> >
> > - if (!list_empty(&tpebs_results)) {
> > + mutex_lock(tpebs_mtx_get());
> > + tpebs_empty = list_empty(&tpebs_results);
> > + mutex_unlock(tpebs_mtx_get());
> > + if (!tpebs_empty) {
> > struct pollfd pollfd = { .events = POLLIN, };
> > int control_fd[2], ack_fd[2], len;
> > char ack_buf[8];
> > @@ -436,8 +467,10 @@ int evsel__tpebs_read(struct evsel *evsel, int cpu_map_idx, int thread)
> > */
> > tpebs_stop();
> >
> > + mutex_lock(tpebs_mtx_get());
> > t = tpebs_retire_lat__find(evsel);
> > val = rint(t->val);
> > + mutex_unlock(tpebs_mtx_get());
> >
> > if (old_count) {
> > count->val = old_count->val + val;
> > @@ -460,9 +493,13 @@ int evsel__tpebs_read(struct evsel *evsel, int cpu_map_idx, int thread)
> > */
> > void evsel__tpebs_close(struct evsel *evsel)
> > {
> > - struct tpebs_retire_lat *t = tpebs_retire_lat__find(evsel);
> > + struct tpebs_retire_lat *t;
> >
> > + mutex_lock(tpebs_mtx_get());
> > + t = tpebs_retire_lat__find(evsel);
> > + list_del_init(&t->nd);
> > tpebs_retire_lat__delete(t);
> > + mutex_unlock(tpebs_mtx_get());
> >
> > if (list_empty(&tpebs_results))
> > tpebs_stop();
>
Powered by blists - more mailing lists