perf hists browser: Warn about lost events

Just like the old perf top --tui and the --stdio version.

But because we have the initial menu to choose which event to show in a
session with multiple events we can see how many chunks were lost in
each of the event types, clarifying which events are being affected the
most.

Cc: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Link: http://lkml.kernel.org/n/tip-47yyqbubmjzch2chezmb21m6@git.kernel.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
This commit is contained in:
Arnaldo Carvalho de Melo
2011-10-29 12:15:04 -02:00
parent 1ca4ff41a3
commit 7b27509fc6
7 changed files with 112 additions and 29 deletions

View File

@@ -738,10 +738,27 @@ static int perf_session_deliver_event(struct perf_session *session,
dump_event(session, event, file_offset, sample);
evsel = perf_evlist__id2evsel(session->evlist, sample->id);
if (evsel != NULL && event->header.type != PERF_RECORD_SAMPLE) {
/*
* XXX We're leaving PERF_RECORD_SAMPLE unnacounted here
* because the tools right now may apply filters, discarding
* some of the samples. For consistency, in the future we
* should have something like nr_filtered_samples and remove
* the sample->period from total_sample_period, etc, KISS for
* now tho.
*
* Also testing against NULL allows us to handle files without
* attr.sample_id_all and/or without PERF_SAMPLE_ID. In the
* future probably it'll be a good idea to restrict event
* processing via perf_session to files with both set.
*/
hists__inc_nr_events(&evsel->hists, event->header.type);
}
switch (event->header.type) {
case PERF_RECORD_SAMPLE:
dump_sample(session, event, sample);
evsel = perf_evlist__id2evsel(session->evlist, sample->id);
if (evsel == NULL) {
++session->hists.stats.nr_unknown_id;
return -1;
@@ -874,11 +891,11 @@ static void perf_session__warn_about_errors(const struct perf_session *session,
const struct perf_event_ops *ops)
{
if (ops->lost == perf_event__process_lost &&
session->hists.stats.total_lost != 0) {
ui__warning("Processed %" PRIu64 " events and LOST %" PRIu64
"!\n\nCheck IO/CPU overload!\n\n",
session->hists.stats.total_period,
session->hists.stats.total_lost);
session->hists.stats.nr_events[PERF_RECORD_LOST] != 0) {
ui__warning("Processed %d events and lost %d chunks!\n\n"
"Check IO/CPU overload!\n\n",
session->hists.stats.nr_events[0],
session->hists.stats.nr_events[PERF_RECORD_LOST]);
}
if (session->hists.stats.nr_unknown_events != 0) {