[PATCH v2] perf record: add dummy event during system wide synthesis

[Ian Rogers]

During the processing of /proc during event synthesis new processes may
start. Add a dummy event if /proc is to be processed, to capture mmaps
for starting processes. This reuses the existing logic for
initial-delay.

v2 fixes the dummy event configuration and a branch stack issue.

Suggested-by: Stephane Eranian <eranian@xxxxxxxxxx>
Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>
---
 tools/perf/builtin-record.c | 19 ++++++++++++++-----
 tools/perf/util/evsel.c     |  5 ++++-
 2 files changed, 18 insertions(+), 6 deletions(-)

diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
index 1ab349abe904..8d1e93351298 100644
--- a/tools/perf/builtin-record.c
+++ b/tools/perf/builtin-record.c
@@ -805,19 +805,28 @@ static int record__open(struct record *rec)
 	int rc = 0;
 
 	/*
-	 * For initial_delay we need to add a dummy event so that we can track
-	 * PERF_RECORD_MMAP while we wait for the initial delay to enable the
-	 * real events, the ones asked by the user.
+	 * For initial_delay or system wide, we need to add a dummy event so
+	 * that we can track PERF_RECORD_MMAP to cover the delay of waiting or
+	 * event synthesis.
 	 */
-	if (opts->initial_delay) {
+	if (opts->initial_delay || target__has_cpu(&opts->target)) {
 		if (perf_evlist__add_dummy(evlist))
 			return -ENOMEM;
 
+		/* Disable tracking of mmaps on lead event. */
 		pos = evlist__first(evlist);
 		pos->tracking = 0;
+		/* Set up dummy event. */
 		pos = evlist__last(evlist);
 		pos->tracking = 1;
-		pos->core.attr.enable_on_exec = 1;
+		/*
+		 * Enable the dummy event when the process is forked for
+		 * initial_delay, immediately for system wide.
+		 */
+		if (opts->initial_delay)
+			pos->core.attr.enable_on_exec = 1;
+		else
+			pos->immediate = 1;
 	}
 
 	perf_evlist__config(evlist, opts, &callchain_param);
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 6a571d322bb2..ca8f9533d8f9 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -1163,11 +1163,14 @@ void perf_evsel__config(struct evsel *evsel, struct record_opts *opts,
 	}
 
 	/*
+	 * A dummy event never triggers any actual counter and therefore
+	 * cannot be used with branch_stack.
+	 *
 	 * For initial_delay, a dummy event is added implicitly.
 	 * The software event will trigger -EOPNOTSUPP error out,
 	 * if BRANCH_STACK bit is set.
 	 */
-	if (opts->initial_delay && is_dummy_event(evsel))
+	if (is_dummy_event(evsel))
 		perf_evsel__reset_sample_bit(evsel, BRANCH_STACK);
 }
 
-- 
2.26.1.301.g55bc3eb7cb9-goog