[PATCH 1/3] perf callchain: Convert children list to rbtree
From: Namhyung Kim
Date: Tue Sep 10 2013 - 04:26:53 EST
From: Namhyung Kim <namhyung.kim@xxxxxxx>
Current collapse stage has a scalability problem which can be
reproduced easily with parallel kernel build. This is because it
needs to traverse every children of callchain linearly during the
collapse/merge stage. Convert it to rbtree reduced the overhead
significantly.
On my 400MB perf.data file which recorded with make -j32 kernel build:
$ time perf --no-pager report --stdio > /dev/null
before:
real 6m22.073s
user 6m18.683s
sys 0m0.706s
after:
real 0m20.780s
user 0m19.962s
sys 0m0.689s
During the perf report the overhead on append_chain_children went down
from 96.69% to 18.16%:
- 18.16% perf perf [.] append_chain_children
- append_chain_children
- 77.48% append_chain_children
+ 69.79% merge_chain_branch
- 22.96% append_chain_children
+ 67.44% merge_chain_branch
+ 30.15% append_chain_children
+ 2.41% callchain_append
+ 7.25% callchain_append
+ 12.26% callchain_append
+ 10.22% merge_chain_branch
+ 11.58% perf perf [.] dso__find_symbol
+ 8.02% perf perf [.] sort__comm_cmp
+ 5.48% perf libc-2.17.so [.] malloc_consolidate
Reported-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Cc: Frederic Weisbecker <fweisbec@xxxxxxxxx>
Signed-off-by: Namhyung Kim <namhyung@xxxxxxxxxx>
---
tools/perf/util/callchain.c | 151 +++++++++++++++++++++++++++++++++-----------
tools/perf/util/callchain.h | 5 +-
2 files changed, 116 insertions(+), 40 deletions(-)
diff --git a/tools/perf/util/callchain.c b/tools/perf/util/callchain.c
index 482f68081cd8..4ac385515c2f 100644
--- a/tools/perf/util/callchain.c
+++ b/tools/perf/util/callchain.c
@@ -21,12 +21,6 @@
__thread struct callchain_cursor callchain_cursor;
-#define chain_for_each_child(child, parent) \
- list_for_each_entry(child, &parent->children, siblings)
-
-#define chain_for_each_child_safe(child, next, parent) \
- list_for_each_entry_safe(child, next, &parent->children, siblings)
-
static void
rb_insert_callchain(struct rb_root *root, struct callchain_node *chain,
enum chain_mode mode)
@@ -71,10 +65,17 @@ static void
__sort_chain_flat(struct rb_root *rb_root, struct callchain_node *node,
u64 min_hit)
{
+ struct rb_node *n;
struct callchain_node *child;
- chain_for_each_child(child, node)
+ n = rb_first(&node->rb_root_in);
+ while (n) {
+ child = rb_entry(n, struct callchain_node, rb_node);
+ n = rb_next(n);
+ rb_erase(&child->rb_node, &node->rb_root_in);
+
__sort_chain_flat(rb_root, child, min_hit);
+ }
if (node->hit && node->hit >= min_hit)
rb_insert_callchain(rb_root, node, CHAIN_FLAT);
@@ -94,11 +95,17 @@ sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root,
static void __sort_chain_graph_abs(struct callchain_node *node,
u64 min_hit)
{
+ struct rb_node *n;
struct callchain_node *child;
node->rb_root = RB_ROOT;
+ n = rb_first(&node->rb_root_in);
+
+ while (n) {
+ child = rb_entry(n, struct callchain_node, rb_node);
+ n = rb_next(n);
+ rb_erase(&child->rb_node, &node->rb_root_in);
- chain_for_each_child(child, node) {
__sort_chain_graph_abs(child, min_hit);
if (callchain_cumul_hits(child) >= min_hit)
rb_insert_callchain(&node->rb_root, child,
@@ -117,13 +124,19 @@ sort_chain_graph_abs(struct rb_root *rb_root, struct callchain_root *chain_root,
static void __sort_chain_graph_rel(struct callchain_node *node,
double min_percent)
{
+ struct rb_node *n;
struct callchain_node *child;
u64 min_hit;
node->rb_root = RB_ROOT;
min_hit = ceil(node->children_hit * min_percent);
- chain_for_each_child(child, node) {
+ n = rb_first(&node->rb_root_in);
+ while (n) {
+ child = rb_entry(n, struct callchain_node, rb_node);
+ n = rb_next(n);
+ rb_erase(&child->rb_node, &node->rb_root_in);
+
__sort_chain_graph_rel(child, min_percent);
if (callchain_cumul_hits(child) >= min_hit)
rb_insert_callchain(&node->rb_root, child,
@@ -173,19 +186,22 @@ create_child(struct callchain_node *parent, bool inherit_children)
return NULL;
}
new->parent = parent;
- INIT_LIST_HEAD(&new->children);
INIT_LIST_HEAD(&new->val);
if (inherit_children) {
- struct callchain_node *next;
+ struct rb_node *n;
+ struct callchain_node *child;
- list_splice(&parent->children, &new->children);
- INIT_LIST_HEAD(&parent->children);
+ new->rb_root_in = parent->rb_root_in;
+ parent->rb_root_in = RB_ROOT;
- chain_for_each_child(next, new)
- next->parent = new;
+ n = rb_first(&new->rb_root_in);
+ while (n) {
+ child = rb_entry(n, struct callchain_node, rb_node);
+ child->parent = new;
+ n = rb_next(n);
+ }
}
- list_add_tail(&new->siblings, &parent->children);
return new;
}
@@ -223,7 +239,7 @@ fill_node(struct callchain_node *node, struct callchain_cursor *cursor)
}
}
-static void
+static struct callchain_node *
add_child(struct callchain_node *parent,
struct callchain_cursor *cursor,
u64 period)
@@ -235,6 +251,19 @@ add_child(struct callchain_node *parent,
new->children_hit = 0;
new->hit = period;
+ return new;
+}
+
+static s64 match_chain(struct callchain_cursor_node *node,
+ struct callchain_list *cnode)
+{
+ struct symbol *sym = node->sym;
+
+ if (cnode->ms.sym && sym &&
+ callchain_param.key == CCKEY_FUNCTION)
+ return cnode->ms.sym->start - sym->start;
+ else
+ return cnode->ip - node->ip;
}
/*
@@ -254,6 +283,12 @@ split_add_child(struct callchain_node *parent,
/* split */
new = create_child(parent, true);
+ /* move existing children under new child */
+ new->rb_root_in = parent->rb_root_in;
+ parent->rb_root_in = RB_ROOT;
+ /* make it the first child */
+ rb_link_node(&new->rb_node, NULL, &parent->rb_root_in.rb_node);
+ rb_insert_color(&new->rb_node, &parent->rb_root_in);
/* split the callchain and move a part to the new child */
old_tail = parent->val.prev;
@@ -272,9 +307,33 @@ split_add_child(struct callchain_node *parent,
/* create a new child for the new branch if any */
if (idx_total < cursor->nr) {
+ struct callchain_node *first;
+ struct callchain_list *cnode;
+ struct callchain_cursor_node *node;
+ struct rb_node *p, **pp;
+
parent->hit = 0;
- add_child(parent, cursor, period);
parent->children_hit += period;
+
+ node = callchain_cursor_current(cursor);
+ new = add_child(parent, cursor, period);
+
+ /*
+ * This is second child since we moved parent's children
+ * to new (first) child above.
+ */
+ p = parent->rb_root_in.rb_node;
+ first = rb_entry(p, struct callchain_node, rb_node);
+ cnode = list_first_entry(&first->val, struct callchain_list,
+ list);
+
+ if (match_chain(node, cnode) < 0)
+ pp = &p->rb_left;
+ else
+ pp = &p->rb_right;
+
+ rb_link_node(&new->rb_node, p, pp);
+ rb_insert_color(&new->rb_node, &parent->rb_root_in);
} else {
parent->hit = period;
}
@@ -291,16 +350,39 @@ append_chain_children(struct callchain_node *root,
u64 period)
{
struct callchain_node *rnode;
+ struct callchain_cursor_node *node;
+ struct rb_node **p = &root->rb_root_in.rb_node;
+ struct rb_node *parent = NULL;
+
+ node = callchain_cursor_current(cursor);
+ if (!node)
+ return;
/* lookup in childrens */
- chain_for_each_child(rnode, root) {
- unsigned int ret = append_chain(rnode, cursor, period);
+ while (*p) {
+ s64 ret;
+ struct callchain_list *cnode;
- if (!ret)
+ parent = *p;
+ rnode = rb_entry(parent, struct callchain_node, rb_node);
+ cnode = list_first_entry(&rnode->val, struct callchain_list,
+ list);
+
+ /* just check first entry */
+ ret = match_chain(node, cnode);
+ if (ret < 0)
+ p = &parent->rb_left;
+ else if (ret > 0)
+ p = &parent->rb_right;
+ else {
+ append_chain(rnode, cursor, period);
goto inc_children_hit;
+ }
}
/* nothing in children, add to the current node */
- add_child(root, cursor, period);
+ rnode = add_child(root, cursor, period);
+ rb_link_node(&rnode->rb_node, parent, p);
+ rb_insert_color(&rnode->rb_node, &root->rb_root_in);
inc_children_hit:
root->children_hit += period;
@@ -325,28 +407,20 @@ append_chain(struct callchain_node *root,
*/
list_for_each_entry(cnode, &root->val, list) {
struct callchain_cursor_node *node;
- struct symbol *sym;
node = callchain_cursor_current(cursor);
if (!node)
break;
- sym = node->sym;
-
- if (cnode->ms.sym && sym &&
- callchain_param.key == CCKEY_FUNCTION) {
- if (cnode->ms.sym->start != sym->start)
- break;
- } else if (cnode->ip != node->ip)
+ if (match_chain(node, cnode) != 0)
break;
- if (!found)
- found = true;
+ found = true;
callchain_cursor_advance(cursor);
}
- /* matches not, relay on the parent */
+ /* matches not, relay no the parent */
if (!found) {
cursor->curr = curr_snap;
cursor->pos = start;
@@ -395,8 +469,9 @@ merge_chain_branch(struct callchain_cursor *cursor,
struct callchain_node *dst, struct callchain_node *src)
{
struct callchain_cursor_node **old_last = cursor->last;
- struct callchain_node *child, *next_child;
+ struct callchain_node *child;
struct callchain_list *list, *next_list;
+ struct rb_node *n;
int old_pos = cursor->nr;
int err = 0;
@@ -412,12 +487,16 @@ merge_chain_branch(struct callchain_cursor *cursor,
append_chain_children(dst, cursor, src->hit);
}
- chain_for_each_child_safe(child, next_child, src) {
+ n = rb_first(&src->rb_root_in);
+ while (n) {
+ child = container_of(n, struct callchain_node, rb_node);
+ n = rb_next(n);
+ rb_erase(&child->rb_node, &src->rb_root_in);
+
err = merge_chain_branch(cursor, dst, child);
if (err)
break;
- list_del(&child->siblings);
free(child);
}
diff --git a/tools/perf/util/callchain.h b/tools/perf/util/callchain.h
index 2b585bc308cf..1b22b6269213 100644
--- a/tools/perf/util/callchain.h
+++ b/tools/perf/util/callchain.h
@@ -21,10 +21,9 @@ enum chain_order {
struct callchain_node {
struct callchain_node *parent;
- struct list_head siblings;
- struct list_head children;
struct list_head val;
struct rb_node rb_node; /* to sort nodes in an rbtree */
+ struct rb_root rb_root_in; /* sorted tree of children */
struct rb_root rb_root; /* sorted tree of children */
unsigned int val_nr;
u64 hit;
@@ -86,8 +85,6 @@ extern __thread struct callchain_cursor callchain_cursor;
static inline void callchain_init(struct callchain_root *root)
{
- INIT_LIST_HEAD(&root->node.siblings);
- INIT_LIST_HEAD(&root->node.children);
INIT_LIST_HEAD(&root->node.val);
root->node.parent = NULL;
--
1.7.11.7
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/