[PATCH 1/5] perf header: do_read shouldn't die

[PATCH 1/5] perf header: do_read shouldn't die

[Arnaldo Carvalho de Melo]

From: Arnaldo Carvalho de Melo <acme@redhat.com>

Propagate the errors instead, its callers already propagate other
errors.

Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/util/header.c |   35 +++++++++++++++++++++--------------
 1 files changed, 21 insertions(+), 14 deletions(-)

diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index df237c3..6b3cb94 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -432,19 +432,19 @@ int perf_header__write(struct perf_header *self, int fd, bool at_exit)
 	return 0;
 }
 
-static void do_read(int fd, void *buf, size_t size)
+static int do_read(int fd, void *buf, size_t size)
 {
 	while (size) {
 		int ret = read(fd, buf, size);
 
-		if (ret < 0)
-			die("failed to read");
-		if (ret == 0)
-			die("failed to read: missing data");
+		if (ret <= 0)
+			return -1;
 
 		size -= ret;
 		buf += ret;
 	}
+
+	return 0;
 }
 
 int perf_header__process_sections(struct perf_header *self, int fd,
@@ -455,7 +455,7 @@ int perf_header__process_sections(struct perf_header *self, int fd,
 	int nr_sections;
 	int sec_size;
 	int idx = 0;
-	int err = 0, feat = 1;
+	int err = -1, feat = 1;
 
 	nr_sections = bitmap_weight(self->adds_features, HEADER_FEAT_BITS);
 	if (!nr_sections)
@@ -469,8 +469,10 @@ int perf_header__process_sections(struct perf_header *self, int fd,
 
 	lseek(fd, self->data_offset + self->data_size, SEEK_SET);
 
-	do_read(fd, feat_sec, sec_size);
+	if (do_read(fd, feat_sec, sec_size))
+		goto out_free;
 
+	err = 0;
 	while (idx < nr_sections && feat < HEADER_LAST_FEATURE) {
 		if (perf_header__has_feat(self, feat)) {
 			struct perf_file_section *sec = &feat_sec[idx++];
@@ -481,18 +483,18 @@ int perf_header__process_sections(struct perf_header *self, int fd,
 		}
 		++feat;
 	}
-
+out_free:
 	free(feat_sec);
 	return err;
-};
+}
 
 int perf_file_header__read(struct perf_file_header *self,
 			   struct perf_header *ph, int fd)
 {
 	lseek(fd, 0, SEEK_SET);
-	do_read(fd, self, sizeof(*self));
 
-	if (self->magic     != PERF_MAGIC ||
+	if (do_read(fd, self, sizeof(*self)) ||
+	    self->magic     != PERF_MAGIC ||
 	    self->attr_size != sizeof(struct perf_file_attr))
 		return -1;
 
@@ -558,7 +560,8 @@ int perf_header__read(struct perf_header *self, int fd)
 		struct perf_header_attr *attr;
 		off_t tmp;
 
-		do_read(fd, &f_attr, sizeof(f_attr));
+		if (do_read(fd, &f_attr, sizeof(f_attr)))
+			goto out_errno;
 		tmp = lseek(fd, 0, SEEK_CUR);
 
 		attr = perf_header_attr__new(&f_attr.attr);
@@ -569,7 +572,8 @@ int perf_header__read(struct perf_header *self, int fd)
 		lseek(fd, f_attr.ids.offset, SEEK_SET);
 
 		for (j = 0; j < nr_ids; j++) {
-			do_read(fd, &f_id, sizeof(f_id));
+			if (do_read(fd, &f_id, sizeof(f_id)))
+				goto out_errno;
 
 			if (perf_header_attr__add_id(attr, f_id) < 0) {
 				perf_header_attr__delete(attr);
@@ -589,7 +593,8 @@ int perf_header__read(struct perf_header *self, int fd)
 		events = malloc(f_header.event_types.size);
 		if (events == NULL)
 			return -ENOMEM;
-		do_read(fd, events, f_header.event_types.size);
+		if (do_read(fd, events, f_header.event_types.size))
+			goto out_errno;
 		event_count =  f_header.event_types.size / sizeof(struct perf_trace_event_type);
 	}
 
@@ -599,6 +604,8 @@ int perf_header__read(struct perf_header *self, int fd)
 
 	self->frozen = 1;
 	return 0;
+out_errno:
+	return -errno;
 }
 
 u64 perf_header__sample_type(struct perf_header *header)
-- 
1.5.5.1

[PATCH 2/5] perf header: perf_header__push_event shouldn't die

[Arnaldo Carvalho de Melo]

From: Arnaldo Carvalho de Melo <acme@redhat.com>

Just propagate eventual errors.

Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/util/header.c       |   16 ++++++++++------
 tools/perf/util/header.h       |    2 +-
 tools/perf/util/parse-events.c |   18 +++++++++++-------
 3 files changed, 22 insertions(+), 14 deletions(-)

diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c
index 6b3cb94..709e325 100644
--- a/tools/perf/util/header.c
+++ b/tools/perf/util/header.c
@@ -105,24 +105,28 @@ struct perf_trace_event_type {
 static int event_count;
 static struct perf_trace_event_type *events;
 
-void perf_header__push_event(u64 id, const char *name)
+int perf_header__push_event(u64 id, const char *name)
 {
 	if (strlen(name) > MAX_EVENT_NAME)
 		pr_warning("Event %s will be truncated\n", name);
 
 	if (!events) {
 		events = malloc(sizeof(struct perf_trace_event_type));
-		if (!events)
-			die("nomem");
+		if (events == NULL)
+			return -ENOMEM;
 	} else {
-		events = realloc(events, (event_count + 1) * sizeof(struct perf_trace_event_type));
-		if (!events)
-			die("nomem");
+		struct perf_trace_event_type *nevents;
+
+		nevents = realloc(events, (event_count + 1) * sizeof(*events));
+		if (nevents == NULL)
+			return -ENOMEM;
+		events = nevents;
 	}
 	memset(&events[event_count], 0, sizeof(struct perf_trace_event_type));
 	events[event_count].event_id = id;
 	strncpy(events[event_count].name, name, MAX_EVENT_NAME - 1);
 	event_count++;
+	return 0;
 }
 
 char *perf_header__find_event(u64 id)
diff --git a/tools/perf/util/header.h b/tools/perf/util/header.h
index d118d05..2b69aab 100644
--- a/tools/perf/util/header.h
+++ b/tools/perf/util/header.h
@@ -64,7 +64,7 @@ int perf_header__write(struct perf_header *self, int fd, bool at_exit);
 int perf_header__add_attr(struct perf_header *self,
 			  struct perf_header_attr *attr);
 
-void perf_header__push_event(u64 id, const char *name);
+int perf_header__push_event(u64 id, const char *name);
 char *perf_header__find_event(u64 id);
 
 struct perf_header_attr *perf_header_attr__new(struct perf_event_attr *attr);
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index e5bc0fb..625ef5a 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -753,11 +753,11 @@ modifier:
 	return ret;
 }
 
-static void store_event_type(const char *orgname)
+static int store_event_type(const char *orgname)
 {
 	char filename[PATH_MAX], *c;
 	FILE *file;
-	int id;
+	int id, n;
 
 	sprintf(filename, "%s/", debugfs_path);
 	strncat(filename, orgname, strlen(orgname));
@@ -769,11 +769,14 @@ static void store_event_type(const char *orgname)
 
 	file = fopen(filename, "r");
 	if (!file)
-		return;
-	if (fscanf(file, "%i", &id) < 1)
-		die("cannot store event ID");
+		return 0;
+	n = fscanf(file, "%i", &id);
 	fclose(file);
-	perf_header__push_event(id, orgname);
+	if (n < 1) {
+		pr_err("cannot store event ID\n");
+		return -EINVAL;
+	}
+	return perf_header__push_event(id, orgname);
 }
 
 int parse_events(const struct option *opt __used, const char *str, int unset __used)
@@ -782,7 +785,8 @@ int parse_events(const struct option *opt __used, const char *str, int unset __u
 	enum event_result ret;
 
 	if (strchr(str, ':'))
-		store_event_type(str);
+		if (store_event_type(str) < 0)
+			return -1;
 
 	for (;;) {
 		if (nr_counters == MAX_COUNTERS)
-- 
1.5.5.1

[PATCH 3/5] perf report: Add --hide-unresolved/-U command line option

[Arnaldo Carvalho de Melo]

From: Arnaldo Carvalho de Melo <acme@redhat.com>

Useful to match the 'overhead' column in 'perf report' with the 'baseline' one
in 'perf diff'.

Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/builtin-report.c |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)

diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c
index 508934b..4292d7a 100644
--- a/tools/perf/builtin-report.c
+++ b/tools/perf/builtin-report.c
@@ -34,6 +34,7 @@
 static char		const *input_name = "perf.data";
 
 static int		force;
+static bool		hide_unresolved;
 
 static int		show_threads;
 static struct perf_read_values	show_threads_values;
@@ -121,7 +122,7 @@ static int process_sample_event(event_t *event, struct perf_session *session)
 		return -1;
 	}
 
-	if (al.filtered)
+	if (al.filtered || (hide_unresolved && al.sym == NULL))
 		return 0;
 
 	if (perf_session__add_hist_entry(session, &al, data.callchain, data.period)) {
@@ -342,6 +343,8 @@ static const struct option options[] = {
 	OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
 		   "separator for columns, no spaces will be added between "
 		   "columns '.' is reserved."),
+	OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved,
+		    "Only display entries resolved to a symbol"),
 	OPT_END()
 };
 
-- 
1.5.5.1

Re: [PATCH 3/5] perf report: Add --hide-unresolved/-U command line option

[Valdis.Kletnieks]

[-- Attachment #1: Type: text/plain, Size: 240 bytes --]

On Mon, 28 Dec 2009 22:48:34 -0200, Arnaldo Carvalho de Melo said:
> Useful to match the 'overhead' column in 'perf report' with the 'baseline' one
> in 'perf diff'.

Looks like a good one for tools/perf/Documentation/examples.txt ?


[-- Attachment #2: Type: application/pgp-signature, Size: 227 bytes --]

[PATCH 4/5] perf diff: Don't add the period for unresolved symbols

[Arnaldo Carvalho de Melo]

From: Arnaldo Carvalho de Melo <acme@redhat.com>

Since we don't add histograms buckets for them, this way the sum of baselines
should be 100%.

Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/builtin-diff.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/perf/builtin-diff.c b/tools/perf/builtin-diff.c
index 1cbecaf..876a4b9 100644
--- a/tools/perf/builtin-diff.c
+++ b/tools/perf/builtin-diff.c
@@ -51,12 +51,12 @@ static int diff__process_sample_event(event_t *event, struct perf_session *sessi
 		return -1;
 	}
 
-	if (al.filtered)
+	if (al.filtered || al.sym == NULL)
 		return 0;
 
 	event__parse_sample(event, session->sample_type, &data);
 
-	if (al.sym && perf_session__add_hist_entry(session, &al, data.period)) {
+	if (perf_session__add_hist_entry(session, &al, data.period)) {
 		pr_warning("problem incrementing symbol count, skipping event\n");
 		return -1;
 	}
-- 
1.5.5.1

[PATCH 5/5] perf diff: Fix support for all --sort combinations

[Arnaldo Carvalho de Melo]

From: Arnaldo Carvalho de Melo <acme@redhat.com>

When we finish creating the hist_entries we _already_ have them sorted "by
name", in fact by what is in --sort, that is exactly how we can find the pairs
in perf_session__match_hists as 'comm', 'dso' & 'symbol' all are strings we
need to find the matches in the baseline session.

So only do the sort by hits followed by a resort by --sort if we need to find
the position for shwowing the --displacement of hist entries.

Now all these modes work correctly:

Example is a simple 'perf record -f find / > /dev/null' ran twice then followed
by the following commands:

  $ perf diff -f --sort comm
  # Baseline  Delta      Command
  # ........ ..........  .......
  #
       0.00%   +100.00%     find
  $ perf diff -f --sort dso
  # Baseline  Delta           Shared Object
  # ........ ..........  ..................
  #
      59.97%     -0.44%  [kernel]
      21.17%     +0.28%  libc-2.5.so
      18.49%     +0.16%  [ext3]
       0.37%             find
  $ perf diff -f --sort symbol | head -8
  # Baseline  Delta      Symbol
  # ........ ..........  ......
  #
       6.21%     +0.36%  [k] ext3fs_dirhash
       3.43%     +0.41%  [.] __GI_strlen
       3.53%     +0.16%  [k] __kmalloc
       3.17%     +0.49%  [k] system_call
       3.06%     +0.37%  [k] ext3_htree_store_dirent
  $ perf diff -f --sort dso,symbol | head -8
  # Baseline  Delta           Shared Object  Symbol
  # ........ ..........  ..................  ......
  #
       6.21%     +0.36%  [ext3]              [k] ext3fs_dirhash
       3.43%     +0.41%  libc-2.5.so         [.] __GI_strlen
       3.53%     +0.16%  [kernel]            [k] __kmalloc
       3.17%     +0.49%  [kernel]            [k] system_call
       3.06%     +0.37%  [ext3]              [k] ext3_htree_store_dirent
  $

And we don't have to do two expensive resorts in the common, non --displacement
case.

Cc: Frédéric Weisbecker <fweisbec@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
---
 tools/perf/builtin-diff.c |   52 ++++++++++++++++++--------------------------
 1 files changed, 21 insertions(+), 31 deletions(-)

diff --git a/tools/perf/builtin-diff.c b/tools/perf/builtin-diff.c
index 876a4b9..924bfb7 100644
--- a/tools/perf/builtin-diff.c
+++ b/tools/perf/builtin-diff.c
@@ -82,29 +82,19 @@ static void perf_session__insert_hist_entry_by_name(struct rb_root *root,
 	struct hist_entry *iter;
 
 	while (*p != NULL) {
-		int cmp;
 		parent = *p;
 		iter = rb_entry(parent, struct hist_entry, rb_node);
-
-		cmp = strcmp(he->map->dso->name, iter->map->dso->name);
-		if (cmp > 0)
+		if (hist_entry__cmp(he, iter) < 0)
 			p = &(*p)->rb_left;
-		else if (cmp < 0)
+		else
 			p = &(*p)->rb_right;
-		else {
-			cmp = strcmp(he->sym->name, iter->sym->name);
-			if (cmp > 0)
-				p = &(*p)->rb_left;
-			else
-				p = &(*p)->rb_right;
-		}
 	}
 
 	rb_link_node(&he->rb_node, parent, p);
 	rb_insert_color(&he->rb_node, root);
 }
 
-static void perf_session__resort_by_name(struct perf_session *self)
+static void perf_session__resort_hist_entries(struct perf_session *self)
 {
 	unsigned long position = 1;
 	struct rb_root tmp = RB_ROOT;
@@ -122,29 +112,28 @@ static void perf_session__resort_by_name(struct perf_session *self)
 	self->hists = tmp;
 }
 
+static void perf_session__set_hist_entries_positions(struct perf_session *self)
+{
+	perf_session__output_resort(self, self->events_stats.total);
+	perf_session__resort_hist_entries(self);
+}
+
 static struct hist_entry *
-perf_session__find_hist_entry_by_name(struct perf_session *self,
-				      struct hist_entry *he)
+perf_session__find_hist_entry(struct perf_session *self,
+			      struct hist_entry *he)
 {
 	struct rb_node *n = self->hists.rb_node;
 
 	while (n) {
 		struct hist_entry *iter = rb_entry(n, struct hist_entry, rb_node);
-		int cmp = strcmp(he->map->dso->name, iter->map->dso->name);
+		int64_t cmp = hist_entry__cmp(he, iter);
 
-		if (cmp > 0)
+		if (cmp < 0)
 			n = n->rb_left;
-		else if (cmp < 0)
+		else if (cmp > 0)
 			n = n->rb_right;
-		else {
-			cmp = strcmp(he->sym->name, iter->sym->name);
-			if (cmp > 0)
-				n = n->rb_left;
-			else if (cmp < 0)
-				n = n->rb_right;
-			else
-				return iter;
-		}
+		else 
+			return iter;
 	}
 
 	return NULL;
@@ -155,11 +144,9 @@ static void perf_session__match_hists(struct perf_session *old_session,
 {
 	struct rb_node *nd;
 
-	perf_session__resort_by_name(old_session);
-
 	for (nd = rb_first(&new_session->hists); nd; nd = rb_next(nd)) {
 		struct hist_entry *pos = rb_entry(nd, struct hist_entry, rb_node);
-		pos->pair = perf_session__find_hist_entry_by_name(old_session, pos);
+		pos->pair = perf_session__find_hist_entry(old_session, pos);
 	}
 }
 
@@ -177,9 +164,12 @@ static int __cmd_diff(void)
 		ret = perf_session__process_events(session[i], &event_ops);
 		if (ret)
 			goto out_delete;
-		perf_session__output_resort(session[i], session[i]->events_stats.total);
 	}
 
+	perf_session__output_resort(session[1], session[1]->events_stats.total);
+	if (show_displacement)
+		perf_session__set_hist_entries_positions(session[0]);
+
 	perf_session__match_hists(session[0], session[1]);
 	perf_session__fprintf_hists(session[1], session[0],
 				    show_displacement, stdout);
-- 
1.5.5.1

Re: [PATCH 5/5] perf diff: Fix support for all --sort combinations

[Arnaldo Carvalho de Melo]

Em Mon, Dec 28, 2009 at 10:48:36PM -0200, Arnaldo Carvalho de Melo escreveu:
> From: Arnaldo Carvalho de Melo <acme@redhat.com>
> 
> When we finish creating the hist_entries we _already_ have them sorted "by
> name", in fact by what is in --sort, that is exactly how we can find the pairs
> in perf_session__match_hists as 'comm', 'dso' & 'symbol' all are strings we
> need to find the matches in the baseline session.
> 
> So only do the sort by hits followed by a resort by --sort if we need to find
> the position for shwowing the --displacement of hist entries.
> 
> Now all these modes work correctly:
> 
> Example is a simple 'perf record -f find / > /dev/null' ran twice then followed
> by the following commands:
> 
>   $ perf diff -f --sort comm
>   # Baseline  Delta      Command
>   # ........ ..........  .......
>   #
>        0.00%   +100.00%     find

Dammit, of course this one needs some more love... will work on it
tomorrow, the others, humm, _should_ be right.

Its late, I should be having some rest already, I guess, or whatever
:-)

- Arnaldo