Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 7b27509f authored by Arnaldo Carvalho de Melo's avatar Arnaldo Carvalho de Melo
Browse files

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: default avatarArnaldo Carvalho de Melo <acme@redhat.com>
parent 1ca4ff41

tools/perf/builtin-top.c

+26 −11
Original line number Diff line number Diff line
@@ -89,6 +89,7 @@ static bool vmlinux_warned; 
static bool			inherit				=  false;

static int			realtime_prio			=      0;

static bool			group				=  false;

static bool			sample_id_all_avail		=   true;

static unsigned int		mmap_pages			=    128;



static bool			dump_symtab                     =  false;
@@ -289,11 +290,13 @@ static void print_sym_table(void) 


	printf("%-*.*s\n", win_width, win_width, graph_dotted_line);



	if (top.total_lost_warned != top.session->hists.stats.total_lost) {

		top.total_lost_warned = top.session->hists.stats.total_lost;

		color_fprintf(stdout, PERF_COLOR_RED, "WARNING:");

		printf(" LOST %" PRIu64 " events, Check IO/CPU overload\n",

		       top.total_lost_warned);

	if (top.sym_evsel->hists.stats.nr_lost_warned !=

	    top.sym_evsel->hists.stats.nr_events[PERF_RECORD_LOST]) {

		top.sym_evsel->hists.stats.nr_lost_warned =

			top.sym_evsel->hists.stats.nr_events[PERF_RECORD_LOST];

		color_fprintf(stdout, PERF_COLOR_RED,

			      "WARNING: LOST %d chunks, Check IO/CPU overload",

			      top.sym_evsel->hists.stats.nr_lost_warned);

		++printed;

	}
@@ -671,6 +674,7 @@ static int symbol_filter(struct map *map __used, struct symbol *sym) 
}



static void perf_event__process_sample(const union perf_event *event,

				       struct perf_evsel *evsel,

				       struct perf_sample *sample,

				       struct perf_session *session)

{
@@ -770,12 +774,8 @@ static void perf_event__process_sample(const union perf_event *event, 
	}



	if (al.sym == NULL || !al.sym->ignore) {

		struct perf_evsel *evsel;

		struct hist_entry *he;



		evsel = perf_evlist__id2evsel(top.evlist, sample->id);

		assert(evsel != NULL);



		if ((sort__has_parent || symbol_conf.use_callchain) &&

		    sample->callchain) {

			err = perf_session__resolve_callchain(session, al.thread,
@@ -807,6 +807,7 @@ static void perf_event__process_sample(const union perf_event *event, 
static void perf_session__mmap_read_idx(struct perf_session *self, int idx)

{

	struct perf_sample sample;

	struct perf_evsel *evsel;

	union perf_event *event;

	int ret;
@@ -817,10 +818,16 @@ static void perf_session__mmap_read_idx(struct perf_session *self, int idx) 
			continue;

		}



		evsel = perf_evlist__id2evsel(self->evlist, sample.id);

		assert(evsel != NULL);



		if (event->header.type == PERF_RECORD_SAMPLE)

			perf_event__process_sample(event, &sample, self);

		else

			perf_event__process_sample(event, evsel, &sample, self);

		else if (event->header.type < PERF_RECORD_MAX) {

			hists__inc_nr_events(&evsel->hists, event->header.type);

			perf_event__process(event, &sample, self);

		} else

			++self->hists.stats.nr_unknown_events;

	}

}
@@ -864,6 +871,8 @@ static void start_counters(struct perf_evlist *evlist) 
		attr->mmap = 1;

		attr->comm = 1;

		attr->inherit = inherit;

retry_sample_id:

		attr->sample_id_all = sample_id_all_avail ? 1 : 0;

try_again:

		if (perf_evsel__open(counter, top.evlist->cpus,

				     top.evlist->threads, group,
@@ -873,6 +882,12 @@ static void start_counters(struct perf_evlist *evlist) 
			if (err == EPERM || err == EACCES) {

				ui__error_paranoid();

				goto out_err;

			} else if (err == EINVAL && sample_id_all_avail) {

				/*

				 * Old kernel, no attr->sample_id_type_all field

				 */

				sample_id_all_avail = false;

				goto retry_sample_id;

			}

			/*

			 * If it's cycles then fall back to hrtimer

tools/perf/util/hist.h

+1 −0
Original line number Diff line number Diff line
@@ -28,6 +28,7 @@ struct events_stats { 
	u64 total_lost;

	u64 total_invalid_chains;

	u32 nr_events[PERF_RECORD_HEADER_MAX];

	u32 nr_lost_warned;

	u32 nr_unknown_events;

	u32 nr_invalid_chains;

	u32 nr_unknown_id;

tools/perf/util/session.c

+23 −6
Original line number Diff line number Diff line
@@ -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) {

tools/perf/util/top.h

+0 −1
Original line number Diff line number Diff line
@@ -19,7 +19,6 @@ struct perf_top { 
	u64		   kernel_samples, us_samples;

	u64		   exact_samples;

	u64		   guest_us_samples, guest_kernel_samples;

	u64		   total_lost_warned;

	int		   print_entries, count_filter, delay_secs;

	int		   freq;

	pid_t		   target_pid, target_tid;

tools/perf/util/ui/browser.c

+17 −4
Original line number Diff line number Diff line
@@ -176,15 +176,28 @@ void ui_browser__handle_resize(struct ui_browser *browser) 
	ui_browser__refresh(browser);

}



int ui_browser__warning(struct ui_browser *browser, const char *format, ...)

int ui_browser__warning(struct ui_browser *browser, int timeout,

			const char *format, ...)

{

	va_list args;

	int key;

	char *text;

	int key = 0, err;



	va_start(args, format);

	while ((key = __ui__warning("Warning!", format, args)) == K_RESIZE)

		ui_browser__handle_resize(browser);

	err = vasprintf(&text, format, args);

	va_end(args);



	if (err < 0) {

		va_start(args, format);

		ui_helpline__vpush(format, args);

		va_end(args);

	} else {

		while ((key == ui__question_window("Warning!", text,

						   "Press any key...",

						   timeout)) == K_RESIZE)

			ui_browser__handle_resize(browser);

		free(text);

	}



	return key;

}