Skip to contentSkip to conversation

perf fixes

GIT PULL New Frederic Weisbecker 2 participants 5 messages Jul 8, 2010 perf events

lore
Share & tools
lore
#0 lore raw

Ingo,

Please pull the perf/urgent branch that can be found at:

git://git.kernel.org/pub/scm/linux/kernel/git/frederic/random-tracing.git
perf/urgent

This resurrects flat and graph (absolute fractals) callchains. All tested with and without newt.

As you know there are still things to fix: filtering by parents (probably a regression) and dso multiple callchains (not a regression).

Thanks,
Frederic
---

Frederic Weisbecker (2):
perf: Resurrect flat callchains
perf: Sync callchains with period based hits

 tools/perf/builtin-report.c |    2 +-
 tools/perf/util/callchain.c |   35 ++++++++++++++++++-----------------
 tools/perf/util/callchain.h |    5 ++++-
 3 files changed, 23 insertions(+), 19 deletions(-)
#1 lore raw

PATCH 1/2 perf: Resurrect flat callchains

Initialize the callchain radix tree root correctly.

When we walk through the parents, we must stop after the root, but since it wasn't well initialized, its parent pointer was random.

Also the number of hits was random because uninitialized, hence it was part of the callchain while the root doesn't contain anything.

This fixes segfaults and percentages followed by empty callchains while running:

perf report -g flat

Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Paul Mackerras <[email protected]>
Cc 2.6.31.x-2.6.34.x <[email protected]>
---
 tools/perf/util/callchain.h |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)
tools/perf/util/callchain.h +3 -0
index 1ca73e4..22dbaec 100644
@@ -49,6 +49,9 @@ static inline void callchain_init(struct callchain_node *node)
 	INIT_LIST_HEAD(&node->brothers);
 	INIT_LIST_HEAD(&node->children);
 	INIT_LIST_HEAD(&node->val);
+
+	node->parent = NULL;
+	node->hit = 0;
 }
 
 static inline u64 cumul_hits(struct callchain_node *node)
Signature · 1 line
1.6.2.3
#2 lore raw

PATCH 2/2 perf: Sync callchains with period based hits

In reply to Frederic Weisbecker

Hists have their hits increased by the event period. And this period based counting is the foundation of all the stats in perf report.

But callchains still use the raw number of hits, without taking the period into account. So when we compute the percentage, absolute based percentages are totally broken, and relative ones too in the first parent level. Because we pass the number of events muliplied by their period as the total number of hits to the callchain filtering, while callchains expect this number to be the number of raw hits.

perf report -g graph was simply not working, showing no graph unless the min percent was zero. And even there the percentage of the branches was always 0. And may be fractal filtering was broken on the first branch level too.

flat also was broken, but it was hidden because of other breakages.

Anyway fix this by counting using periods on callchains.

Cc Ingo Molnar <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Arnaldo Carvalho de Melo <[email protected]>
Cc Paul Mackerras <[email protected]>
---
 tools/perf/builtin-report.c |    2 +-
 tools/perf/util/callchain.c |   35 ++++++++++++++++++-----------------
 tools/perf/util/callchain.h |    2 +-
 3 files changed, 20 insertions(+), 19 deletions(-)
tools/perf/builtin-report.c +1 -1
index 3592057..fd7407c 100644
@@ -107,7 +107,7 @@ static int perf_session__add_hist_entry(struct perf_session *self,
 		goto out_free_syms;
 	err = 0;
 	if (symbol_conf.use_callchain) {
-		err = append_chain(he->callchain, data->callchain, syms);
+		err = append_chain(he->callchain, data->callchain, syms, data->period);
 		if (err)
 			goto out_free_syms;
 	}
tools/perf/util/callchain.c +18 -17
index 62b69ad..52c777e 100644
@@ -230,7 +230,7 @@ fill_node(struct callchain_node *node, struct resolved_chain *chain, int start)
 
 static void
 add_child(struct callchain_node *parent, struct resolved_chain *chain,
-	  int start)
+	  int start, u64 period)
 {
 	struct callchain_node *new;
 
@@ -238,7 +238,7 @@ add_child(struct callchain_node *parent, struct resolved_chain *chain,
 	fill_node(new, chain, start);
 
 	new->children_hit = 0;
-	new->hit = 1;
+	new->hit = period;
 }
 
 /*
@@ -248,7 +248,8 @@ add_child(struct callchain_node *parent, struct resolved_chain *chain,
  */
 static void
 split_add_child(struct callchain_node *parent, struct resolved_chain *chain,
-		struct callchain_list *to_split, int idx_parents, int idx_local)
+		struct callchain_list *to_split, int idx_parents, int idx_local,
+		u64 period)
 {
 	struct callchain_node *new;
 	struct list_head *old_tail;
@@ -275,41 +276,41 @@ split_add_child(struct callchain_node *parent, struct resolved_chain *chain,
 	/* create a new child for the new branch if any */
 	if (idx_total < chain->nr) {
 		parent->hit = 0;
-		add_child(parent, chain, idx_total);
-		parent->children_hit++;
+		add_child(parent, chain, idx_total, period);
+		parent->children_hit += period;
 	} else {
-		parent->hit = 1;
+		parent->hit = period;
 	}
 }
 
 static int
 __append_chain(struct callchain_node *root, struct resolved_chain *chain,
-	       unsigned int start);
+	       unsigned int start, u64 period);
 
 static void
 __append_chain_children(struct callchain_node *root,
 			struct resolved_chain *chain,
-			unsigned int start)
+			unsigned int start, u64 period)
 {
 	struct callchain_node *rnode;
 
 	/* lookup in childrens */
 	chain_for_each_child(rnode, root) {
-		unsigned int ret = __append_chain(rnode, chain, start);
+		unsigned int ret = __append_chain(rnode, chain, start, period);
 
 		if (!ret)
 			goto inc_children_hit;
 	}
 	/* nothing in children, add to the current node */
-	add_child(root, chain, start);
+	add_child(root, chain, start, period);
 
 inc_children_hit:
-	root->children_hit++;
+	root->children_hit += period;
 }
 
 static int
 __append_chain(struct callchain_node *root, struct resolved_chain *chain,
-	       unsigned int start)
+	       unsigned int start, u64 period)
 {
 	struct callchain_list *cnode;
 	unsigned int i = start;
@@ -345,18 +346,18 @@ __append_chain(struct callchain_node *root, struct resolved_chain *chain,
 
 	/* we match only a part of the node. Split it and add the new chain */
 	if (i - start < root->val_nr) {
-		split_add_child(root, chain, cnode, start, i - start);
+		split_add_child(root, chain, cnode, start, i - start, period);
 		return 0;
 	}
 
 	/* we match 100% of the path, increment the hit */
 	if (i - start == root->val_nr && i == chain->nr) {
-		root->hit++;
+		root->hit += period;
 		return 0;
 	}
 
 	/* We match the node and still have a part remaining */
-	__append_chain_children(root, chain, i);
+	__append_chain_children(root, chain, i, period);
 
 	return 0;
 }
@@ -380,7 +381,7 @@ static void filter_context(struct ip_callchain *old, struct resolved_chain *new,
 
 
 int append_chain(struct callchain_node *root, struct ip_callchain *chain,
-		 struct map_symbol *syms)
+		 struct map_symbol *syms, u64 period)
 {
 	struct resolved_chain *filtered;
 
@@ -397,7 +398,7 @@ int append_chain(struct callchain_node *root, struct ip_callchain *chain,
 	if (!filtered->nr)
 		goto end;
 
-	__append_chain_children(root, filtered, 0);
+	__append_chain_children(root, filtered, 0, period);
 end:
 	free(filtered);
 
tools/perf/util/callchain.h +1 -1
index 22dbaec..f2e9ee1 100644
@@ -61,7 +61,7 @@ static inline u64 cumul_hits(struct callchain_node *node)
 
 int register_callchain_param(struct callchain_param *param);
 int append_chain(struct callchain_node *root, struct ip_callchain *chain,
-		 struct map_symbol *syms);
+		 struct map_symbol *syms, u64 period);
 
 bool ip_callchain__valid(struct ip_callchain *chain, event_t *event);
 #endif	/* __PERF_CALLCHAIN_H */
Signature · 1 line
1.6.2.3
#3 lore raw
In reply to Frederic Weisbecker
Quoted 13 lines · Frederic Weisbecker

Ingo,

Please pull the perf/urgent branch that can be found at:

git://git.kernel.org/pub/scm/linux/kernel/git/frederic/random-tracing.git
perf/urgent

This resurrects flat and graph (absolute fractals) callchains. All tested with and without newt.

As you know there are still things to fix: filtering by parents (probably a regression) and dso multiple callchains (not a regression).

Quoted 7 lines

Thanks,
Frederic
---

Frederic Weisbecker (2):
perf: Resurrect flat callchains
perf: Sync callchains with period based hits

BTW, the second is not that small for a regression fix. If that's a problem, tell me and I will move this one to perf/core.

#4 lore raw
In reply to Frederic Weisbecker
Quoted 22 lines · Frederic Weisbecker

Ingo,

Please pull the perf/urgent branch that can be found at:

git://git.kernel.org/pub/scm/linux/kernel/git/frederic/random-tracing.git
perf/urgent

This resurrects flat and graph (absolute fractals) callchains. All tested with and without newt.

As you know there are still things to fix: filtering by parents (probably a regression) and dso multiple callchains (not a regression).

Thanks,
Frederic
---

Frederic Weisbecker (2):
perf: Resurrect flat callchains
perf: Sync callchains with period based hits

Quoted 4 lines
 tools/perf/builtin-report.c |    2 +-
 tools/perf/util/callchain.c |   35 ++++++++++++++++++-----------------
 tools/perf/util/callchain.h |    5 ++++-
 3 files changed, 23 insertions(+), 19 deletions(-)

Pulled, thanks Frederic!

Ingo