Merge tag 'perf-urgent-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/g…

…it/acme/linux into perf/urgent Pull perf/urgent fixes from Arnaldo Carvalho de Melo: * Fix parsing with no sample_id_all bit set, this regression prevents perf from reading old perf.data files generated in systems where perf_event_attr.sample_id_all isn't available, from Adrian Hunter. * Add signal checking to the inner 'perf trace' event processing loop, allowing faster response to control+C. * Fix formatting of long symbol names removing the hardcoding of a buffer size used to format histogram entries, which was truncating the lines. * Separate progress bar update when processing events, reducing potentially big overhead in not needed TUI progress bar screen updates, from Jiri Olsa. * Fix 'perf trace' build in architectures where MAP_32BIT is not defined, from Kyle McMartin. Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: Ingo Molnar <mingo@kernel.org>

Merge tag 'perf-urgent-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/g…
…it/acme/linux into perf/urgent Pull perf/urgent fixes from Arnaldo Carvalho de Melo: * Fix parsing with no sample_id_all bit set, this regression prevents perf from reading old perf.data files generated in systems where perf_event_attr.sample_id_all isn't available, from Adrian Hunter. * Add signal checking to the inner 'perf trace' event processing loop, allowing faster response to control+C. * Fix formatting of long symbol names removing the hardcoding of a buffer size used to format histogram entries, which was truncating the lines. * Separate progress bar update when processing events, reducing potentially big overhead in not needed TUI progress bar screen updates, from Jiri Olsa. * Fix 'perf trace' build in architectures where MAP_32BIT is not defined, from Kyle McMartin. Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: Ingo Molnar <mingo@kernel.org>
Ingo Molnar
2 parents 816434ec4a 526fd8d4f7
Showing 8 changed files Side-by-side Diff
tools/perf/Makefile
tools/perf/builtin-trace.c
tools/perf/tests/builtin-test.c
tools/perf/tests/parse-no-sample-id-all.c
tools/perf/tests/tests.h
tools/perf/ui/stdio/hist.c
tools/perf/util/evlist.c
tools/perf/util/session.c
@@ -394,6 +394,8 @@
 LIB_OBJS += $(OUTPUT)tests/perf-time-to-tsc.o
 endif
 LIB_OBJS += $(OUTPUT)tests/code-reading.o
+LIB_OBJS += $(OUTPUT)tests/sample-parsing.o
+LIB_OBJS += $(OUTPUT)tests/parse-no-sample-id-all.o
  
 BUILTIN_OBJS += $(OUTPUT)builtin-annotate.o
 BUILTIN_OBJS += $(OUTPUT)builtin-bench.o
@@ -439,7 +441,6 @@
 ifneq ($(OUTPUT),)
   CFLAGS += -I$(OUTPUT)
 endif
-LIB_OBJS += $(OUTPUT)tests/sample-parsing.o
  
 ifdef NO_LIBELF
 EXTLIBS := $(filter-out -lelf,$(EXTLIBS))
@@ -100,7 +100,9 @@
  
 	P_MMAP_FLAG(SHARED);
 	P_MMAP_FLAG(PRIVATE);
+#ifdef MAP_32BIT
 	P_MMAP_FLAG(32BIT);
+#endif
 	P_MMAP_FLAG(ANONYMOUS);
 	P_MMAP_FLAG(DENYWRITE);
 	P_MMAP_FLAG(EXECUTABLE);
@@ -994,6 +996,9 @@
  
 			handler = evsel->handler.func;
 			handler(trace, evsel, &sample);
+
+			if (done)
+				goto out_unmap_evlist;
 		}
 	}
  
@@ -112,6 +112,10 @@
 		.func = test__keep_tracking,
 	},
 	{
+		.desc = "Test parsing with no sample_id_all bit set",
+		.func = test__parse_no_sample_id_all,
+	},
+	{
 		.func = NULL,
 	},
 };
+#include <sys/types.h>
+#include <stddef.h>
+
+#include "tests.h"
+
+#include "event.h"
+#include "evlist.h"
+#include "header.h"
+#include "util.h"
+
+static int process_event(struct perf_evlist **pevlist, union perf_event *event)
+{
+	struct perf_sample sample;
+
+	if (event->header.type == PERF_RECORD_HEADER_ATTR) {
+		if (perf_event__process_attr(NULL, event, pevlist)) {
+			pr_debug("perf_event__process_attr failed\n");
+			return -1;
+		}
+		return 0;
+	}
+
+	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
+		return -1;
+
+	if (!*pevlist)
+		return -1;
+
+	if (perf_evlist__parse_sample(*pevlist, event, &sample)) {
+		pr_debug("perf_evlist__parse_sample failed\n");
+		return -1;
+	}
+
+	return 0;
+}
+
+static int process_events(union perf_event **events, size_t count)
+{
+	struct perf_evlist *evlist = NULL;
+	int err = 0;
+	size_t i;
+
+	for (i = 0; i < count && !err; i++)
+		err = process_event(&evlist, events[i]);
+
+	if (evlist)
+		perf_evlist__delete(evlist);
+
+	return err;
+}
+
+struct test_attr_event {
+	struct attr_event attr;
+	u64 id;
+};
+
+/**
+ * test__parse_no_sample_id_all - test parsing with no sample_id_all bit set.
+ *
+ * This function tests parsing data produced on kernel's that do not support the
+ * sample_id_all bit.  Without the sample_id_all bit, non-sample events (such as
+ * mmap events) do not have an id sample appended, and consequently logic
+ * designed to determine the id will not work.  That case happens when there is
+ * more than one selected event, so this test processes three events: 2
+ * attributes representing the selected events and one mmap event.
+ *
+ * Return: %0 on success, %-1 if the test fails.
+ */
+int test__parse_no_sample_id_all(void)
+{
+	int err;
+
+	struct test_attr_event event1 = {
+		.attr = {
+			.header = {
+				.type = PERF_RECORD_HEADER_ATTR,
+				.size = sizeof(struct test_attr_event),
+			},
+		},
+		.id = 1,
+	};
+	struct test_attr_event event2 = {
+		.attr = {
+			.header = {
+				.type = PERF_RECORD_HEADER_ATTR,
+				.size = sizeof(struct test_attr_event),
+			},
+		},
+		.id = 2,
+	};
+	struct mmap_event event3 = {
+		.header = {
+			.type = PERF_RECORD_MMAP,
+			.size = sizeof(struct mmap_event),
+		},
+	};
+	union perf_event *events[] = {
+		(union perf_event *)&event1,
+		(union perf_event *)&event2,
+		(union perf_event *)&event3,
+	};
+
+	err = process_events(events, ARRAY_SIZE(events));
+	if (err)
+		return -1;
+
+	return 0;
+}
@@ -39,6 +39,7 @@
 int test__code_reading(void);
 int test__sample_parsing(void);
 int test__keep_tracking(void);
+int test__parse_no_sample_id_all(void);
  
 #endif /* TESTS_H */
@@ -350,9 +350,9 @@
 }
  
 static int hist_entry__fprintf(struct hist_entry *he, size_t size,
-			       struct hists *hists, FILE *fp)
+			       struct hists *hists,
+			       char *bf, size_t bfsz, FILE *fp)
 {
-	char bf[512];
 	int ret;
 	struct perf_hpp hpp = {
 		.buf		= bf,
@@ -360,8 +360,8 @@
 	};
 	bool color = !symbol_conf.field_sep;
  
-	if (size == 0 || size > sizeof(bf))
-		size = hpp.size = sizeof(bf);
+	if (size == 0 || size > bfsz)
+		size = hpp.size = bfsz;
  
 	ret = hist_entry__period_snprintf(&hpp, he, color);
 	hist_entry__sort_snprintf(he, bf + ret, size - ret, hists);
@@ -392,6 +392,8 @@
 		.ptr	= hists_to_evsel(hists),
 	};
 	bool first = true;
+	size_t linesz;
+	char *line = NULL;
  
 	init_rem_hits();
  
@@ -479,6 +481,13 @@
 		goto out;
  
 print_entries:
+	linesz = hists__sort_list_width(hists) + 3 + 1;
+	line = malloc(linesz);
+	if (line == NULL) {
+		ret = -1;
+		goto out;
+	}
+
 	for (nd = rb_first(&hists->entries); nd; nd = rb_next(nd)) {
 		struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
 		float percent = h->stat.period * 100.0 /
  
@@ -490,10 +499,10 @@
 		if (percent < min_pcnt)
 			continue;
  
-		ret += hist_entry__fprintf(h, max_cols, hists, fp);
+		ret += hist_entry__fprintf(h, max_cols, hists, line, linesz, fp);
  
 		if (max_rows && ++nr_rows >= max_rows)
-			goto out;
+			break;
  
 		if (h->ms.map == NULL && verbose > 1) {
 			__map_groups__fprintf_maps(&h->thread->mg,
@@ -501,6 +510,8 @@
 			fprintf(fp, "%.10s end\n", graph_dotted_line);
 		}
 	}
+
+	free(line);
 out:
 	free(rem_sq_bracket);
  
@@ -446,20 +446,25 @@
 static struct perf_evsel *perf_evlist__event2evsel(struct perf_evlist *evlist,
 						   union perf_event *event)
 {
+	struct perf_evsel *first = perf_evlist__first(evlist);
 	struct hlist_head *head;
 	struct perf_sample_id *sid;
 	int hash;
 	u64 id;
  
 	if (evlist->nr_entries == 1)
-		return perf_evlist__first(evlist);
+		return first;
  
+	if (!first->attr.sample_id_all &&
+	    event->header.type != PERF_RECORD_SAMPLE)
+		return first;
+
 	if (perf_evlist__event2id(evlist, event, &id))
 		return NULL;
  
 	/* Synthesized events have an id of zero */
 	if (!id)
-		return perf_evlist__first(evlist);
+		return first;
  
 	hash = hash_64(id, PERF_EVLIST__HLIST_BITS);
 	head = &evlist->heads[hash];
@@ -504,6 +504,7 @@
 	u64 limit = os->next_flush;
 	u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL;
 	unsigned idx = 0, progress_next = os->nr_samples / 16;
+	bool show_progress = limit == ULLONG_MAX;
 	int ret;
  
 	if (!tool->ordered_samples || !limit)
@@ -526,7 +527,7 @@
 		os->last_flush = iter->timestamp;
 		list_del(&iter->list);
 		list_add(&iter->list, &os->sample_cache);
-		if (++idx >= progress_next) {
+		if (show_progress && (++idx >= progress_next)) {
 			progress_next += os->nr_samples / 16;
 			ui_progress__update(idx, os->nr_samples,
 					    "Processing time ordered events...");
...	...	@@ -394,6 +394,8 @@
394	394	LIB_OBJS += $(OUTPUT)tests/perf-time-to-tsc.o
395	395	endif
396	396	LIB_OBJS += $(OUTPUT)tests/code-reading.o
	397	+LIB_OBJS += $(OUTPUT)tests/sample-parsing.o
	398	+LIB_OBJS += $(OUTPUT)tests/parse-no-sample-id-all.o
397	399
398	400	BUILTIN_OBJS += $(OUTPUT)builtin-annotate.o
399	401	BUILTIN_OBJS += $(OUTPUT)builtin-bench.o
...	...	@@ -439,7 +441,6 @@
439	441	ifneq ($(OUTPUT),)
440	442	CFLAGS += -I$(OUTPUT)
441	443	endif
442		-LIB_OBJS += $(OUTPUT)tests/sample-parsing.o
443	444
444	445	ifdef NO_LIBELF
445	446	EXTLIBS := $(filter-out -lelf,$(EXTLIBS))
...	...	@@ -100,7 +100,9 @@
100	100
101	101	P_MMAP_FLAG(SHARED);
102	102	P_MMAP_FLAG(PRIVATE);
	103	+#ifdef MAP_32BIT
103	104	P_MMAP_FLAG(32BIT);
	105	+#endif
104	106	P_MMAP_FLAG(ANONYMOUS);
105	107	P_MMAP_FLAG(DENYWRITE);
106	108	P_MMAP_FLAG(EXECUTABLE);
...	...	@@ -994,6 +996,9 @@
994	996
995	997	handler = evsel->handler.func;
996	998	handler(trace, evsel, &sample);
	999	+
	1000	+ if (done)
	1001	+ goto out_unmap_evlist;
997	1002	}
998	1003	}
999	1004
...	...	@@ -112,6 +112,10 @@
112	112	.func = test__keep_tracking,
113	113	},
114	114	{
	115	+ .desc = "Test parsing with no sample_id_all bit set",
	116	+ .func = test__parse_no_sample_id_all,
	117	+ },
	118	+ {
115	119	.func = NULL,
116	120	},
117	121	};
	1	+#include <sys/types.h>
	2	+#include <stddef.h>
	3	+
	4	+#include "tests.h"
	5	+
	6	+#include "event.h"
	7	+#include "evlist.h"
	8	+#include "header.h"
	9	+#include "util.h"
	10	+
	11	+static int process_event(struct perf_evlist *pevlist, union perf_event event)
	12	+{
	13	+ struct perf_sample sample;
	14	+
	15	+ if (event->header.type == PERF_RECORD_HEADER_ATTR) {
	16	+ if (perf_event__process_attr(NULL, event, pevlist)) {
	17	+ pr_debug("perf_event__process_attr failed\n");
	18	+ return -1;
	19	+ }
	20	+ return 0;
	21	+ }
	22	+
	23	+ if (event->header.type >= PERF_RECORD_USER_TYPE_START)
	24	+ return -1;
	25	+
	26	+ if (!*pevlist)
	27	+ return -1;
	28	+
	29	+ if (perf_evlist__parse_sample(*pevlist, event, &sample)) {
	30	+ pr_debug("perf_evlist__parse_sample failed\n");
	31	+ return -1;
	32	+ }
	33	+
	34	+ return 0;
	35	+}
	36	+
	37	+static int process_events(union perf_event **events, size_t count)
	38	+{
	39	+ struct perf_evlist *evlist = NULL;
	40	+ int err = 0;
	41	+ size_t i;
	42	+
	43	+ for (i = 0; i < count && !err; i++)
	44	+ err = process_event(&evlist, events[i]);
	45	+
	46	+ if (evlist)
	47	+ perf_evlist__delete(evlist);
	48	+
	49	+ return err;
	50	+}
	51	+
	52	+struct test_attr_event {
	53	+ struct attr_event attr;
	54	+ u64 id;
	55	+};
	56	+
	57	+/**
	58	+ * test__parse_no_sample_id_all - test parsing with no sample_id_all bit set.
	59	+ *
	60	+ * This function tests parsing data produced on kernel's that do not support the
	61	+ * sample_id_all bit. Without the sample_id_all bit, non-sample events (such as
	62	+ * mmap events) do not have an id sample appended, and consequently logic
	63	+ * designed to determine the id will not work. That case happens when there is
	64	+ * more than one selected event, so this test processes three events: 2
	65	+ * attributes representing the selected events and one mmap event.
	66	+ *
	67	+ * Return: %0 on success, %-1 if the test fails.
	68	+ */
	69	+int test__parse_no_sample_id_all(void)
	70	+{
	71	+ int err;
	72	+
	73	+ struct test_attr_event event1 = {
	74	+ .attr = {
	75	+ .header = {
	76	+ .type = PERF_RECORD_HEADER_ATTR,
	77	+ .size = sizeof(struct test_attr_event),
	78	+ },
	79	+ },
	80	+ .id = 1,
	81	+ };
	82	+ struct test_attr_event event2 = {
	83	+ .attr = {
	84	+ .header = {
	85	+ .type = PERF_RECORD_HEADER_ATTR,
	86	+ .size = sizeof(struct test_attr_event),
	87	+ },
	88	+ },
	89	+ .id = 2,
	90	+ };
	91	+ struct mmap_event event3 = {
	92	+ .header = {
	93	+ .type = PERF_RECORD_MMAP,
	94	+ .size = sizeof(struct mmap_event),
	95	+ },
	96	+ };
	97	+ union perf_event *events[] = {
	98	+ (union perf_event *)&event1,
	99	+ (union perf_event *)&event2,
	100	+ (union perf_event *)&event3,
	101	+ };
	102	+
	103	+ err = process_events(events, ARRAY_SIZE(events));
	104	+ if (err)
	105	+ return -1;
	106	+
	107	+ return 0;
	108	+}
...	...	@@ -39,6 +39,7 @@
39	39	int test__code_reading(void);
40	40	int test__sample_parsing(void);
41	41	int test__keep_tracking(void);
	42	+int test__parse_no_sample_id_all(void);
42	43
43	44	#endif /* TESTS_H */
...	...	@@ -350,9 +350,9 @@
350	350	}
351	351
352	352	static int hist_entry__fprintf(struct hist_entry *he, size_t size,
353		- struct hists hists, FILE fp)
	353	+ struct hists *hists,
	354	+ char bf, size_t bfsz, FILE fp)
354	355	{
355		- char bf[512];
356	356	int ret;
357	357	struct perf_hpp hpp = {
358	358	.buf = bf,
...	...	@@ -360,8 +360,8 @@
360	360	};
361	361	bool color = !symbol_conf.field_sep;
362	362
363		- if (size == 0 \|\| size > sizeof(bf))
364		- size = hpp.size = sizeof(bf);
	363	+ if (size == 0 \|\| size > bfsz)
	364	+ size = hpp.size = bfsz;
365	365
366	366	ret = hist_entry__period_snprintf(&hpp, he, color);
367	367	hist_entry__sort_snprintf(he, bf + ret, size - ret, hists);
...	...	@@ -392,6 +392,8 @@
392	392	.ptr = hists_to_evsel(hists),
393	393	};
394	394	bool first = true;
	395	+ size_t linesz;
	396	+ char *line = NULL;
395	397
396	398	init_rem_hits();
397	399
...	...	@@ -479,6 +481,13 @@
479	481	goto out;
480	482
481	483	print_entries:
	484	+ linesz = hists__sort_list_width(hists) + 3 + 1;
	485	+ line = malloc(linesz);
	486	+ if (line == NULL) {
	487	+ ret = -1;
	488	+ goto out;
	489	+ }
	490	+
482	491	for (nd = rb_first(&hists->entries); nd; nd = rb_next(nd)) {
483	492	struct hist_entry *h = rb_entry(nd, struct hist_entry, rb_node);
484	493	float percent = h->stat.period * 100.0 /
485	494
...	...	@@ -490,10 +499,10 @@
490	499	if (percent < min_pcnt)
491	500	continue;
492	501
493		- ret += hist_entry__fprintf(h, max_cols, hists, fp);
	502	+ ret += hist_entry__fprintf(h, max_cols, hists, line, linesz, fp);
494	503
495	504	if (max_rows && ++nr_rows >= max_rows)
496		- goto out;
	505	+ break;
497	506
498	507	if (h->ms.map == NULL && verbose > 1) {
499	508	__map_groups__fprintf_maps(&h->thread->mg,
...	...	@@ -501,6 +510,8 @@
501	510	fprintf(fp, "%.10s end\n", graph_dotted_line);
502	511	}
503	512	}
	513	+
	514	+ free(line);
504	515	out:
505	516	free(rem_sq_bracket);
506	517
...	...	@@ -446,20 +446,25 @@
446	446	static struct perf_evsel perf_evlist__event2evsel(struct perf_evlist evlist,
447	447	union perf_event *event)
448	448	{
	449	+ struct perf_evsel *first = perf_evlist__first(evlist);
449	450	struct hlist_head *head;
450	451	struct perf_sample_id *sid;
451	452	int hash;
452	453	u64 id;
453	454
454	455	if (evlist->nr_entries == 1)
455		- return perf_evlist__first(evlist);
	456	+ return first;
456	457
	458	+ if (!first->attr.sample_id_all &&
	459	+ event->header.type != PERF_RECORD_SAMPLE)
	460	+ return first;
	461	+
457	462	if (perf_evlist__event2id(evlist, event, &id))
458	463	return NULL;
459	464
460	465	/* Synthesized events have an id of zero */
461	466	if (!id)
462		- return perf_evlist__first(evlist);
	467	+ return first;
463	468
464	469	hash = hash_64(id, PERF_EVLIST__HLIST_BITS);
465	470	head = &evlist->heads[hash];
...	...	@@ -504,6 +504,7 @@
504	504	u64 limit = os->next_flush;
505	505	u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL;
506	506	unsigned idx = 0, progress_next = os->nr_samples / 16;
	507	+ bool show_progress = limit == ULLONG_MAX;
507	508	int ret;
508	509
509	510	if (!tool->ordered_samples \|\| !limit)
...	...	@@ -526,7 +527,7 @@
526	527	os->last_flush = iter->timestamp;
527	528	list_del(&iter->list);
528	529	list_add(&iter->list, &os->sample_cache);
529		- if (++idx >= progress_next) {
	530	+ if (show_progress && (++idx >= progress_next)) {
530	531	progress_next += os->nr_samples / 16;
531	532	ui_progress__update(idx, os->nr_samples,
532	533	"Processing time ordered events...");