2009-06-26 22:28:00 +08:00
|
|
|
/*
|
2011-01-14 11:51:58 +08:00
|
|
|
* Copyright (C) 2009-2011, Frederic Weisbecker <fweisbec@gmail.com>
|
2009-06-26 22:28:00 +08:00
|
|
|
*
|
|
|
|
* Handle the callchains from the stream in an ad-hoc radix tree and then
|
|
|
|
* sort them in an rbtree.
|
|
|
|
*
|
2009-07-01 11:35:15 +08:00
|
|
|
* Using a radix for code path provides a fast retrieval and factorizes
|
|
|
|
* memory use. Also that lets us use the paths in a hierarchical graph view.
|
|
|
|
*
|
2009-06-26 22:28:00 +08:00
|
|
|
*/
|
|
|
|
|
|
|
|
#include <stdlib.h>
|
|
|
|
#include <stdio.h>
|
|
|
|
#include <stdbool.h>
|
|
|
|
#include <errno.h>
|
2009-08-09 10:19:15 +08:00
|
|
|
#include <math.h>
|
2009-06-26 22:28:00 +08:00
|
|
|
|
2010-05-20 23:15:33 +08:00
|
|
|
#include "util.h"
|
2009-06-26 22:28:00 +08:00
|
|
|
#include "callchain.h"
|
|
|
|
|
2010-06-04 19:02:07 +08:00
|
|
|
bool ip_callchain__valid(struct ip_callchain *chain, const event_t *event)
|
2010-05-09 22:47:13 +08:00
|
|
|
{
|
|
|
|
unsigned int chain_size = event->header.size;
|
|
|
|
chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
|
|
|
|
return chain->nr * sizeof(u64) <= chain_size;
|
|
|
|
}
|
|
|
|
|
2009-07-02 23:58:19 +08:00
|
|
|
#define chain_for_each_child(child, parent) \
|
|
|
|
list_for_each_entry(child, &parent->children, brothers)
|
|
|
|
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
#define chain_for_each_child_safe(child, next, parent) \
|
|
|
|
list_for_each_entry_safe(child, next, &parent->children, brothers)
|
|
|
|
|
2009-07-01 11:35:15 +08:00
|
|
|
static void
|
2009-07-02 23:58:21 +08:00
|
|
|
rb_insert_callchain(struct rb_root *root, struct callchain_node *chain,
|
|
|
|
enum chain_mode mode)
|
2009-06-26 22:28:00 +08:00
|
|
|
{
|
|
|
|
struct rb_node **p = &root->rb_node;
|
|
|
|
struct rb_node *parent = NULL;
|
|
|
|
struct callchain_node *rnode;
|
2011-01-14 11:51:59 +08:00
|
|
|
u64 chain_cumul = callchain_cumul_hits(chain);
|
2009-06-26 22:28:00 +08:00
|
|
|
|
|
|
|
while (*p) {
|
2009-08-07 13:11:05 +08:00
|
|
|
u64 rnode_cumul;
|
|
|
|
|
2009-06-26 22:28:00 +08:00
|
|
|
parent = *p;
|
|
|
|
rnode = rb_entry(parent, struct callchain_node, rb_node);
|
2011-01-14 11:51:59 +08:00
|
|
|
rnode_cumul = callchain_cumul_hits(rnode);
|
2009-06-26 22:28:00 +08:00
|
|
|
|
2009-07-02 23:58:21 +08:00
|
|
|
switch (mode) {
|
2009-07-05 13:39:21 +08:00
|
|
|
case CHAIN_FLAT:
|
2009-07-02 23:58:21 +08:00
|
|
|
if (rnode->hit < chain->hit)
|
|
|
|
p = &(*p)->rb_left;
|
|
|
|
else
|
|
|
|
p = &(*p)->rb_right;
|
|
|
|
break;
|
2009-07-05 13:39:21 +08:00
|
|
|
case CHAIN_GRAPH_ABS: /* Falldown */
|
|
|
|
case CHAIN_GRAPH_REL:
|
2009-08-07 13:11:05 +08:00
|
|
|
if (rnode_cumul < chain_cumul)
|
2009-07-02 23:58:21 +08:00
|
|
|
p = &(*p)->rb_left;
|
|
|
|
else
|
|
|
|
p = &(*p)->rb_right;
|
|
|
|
break;
|
2009-08-15 18:26:57 +08:00
|
|
|
case CHAIN_NONE:
|
2009-07-02 23:58:21 +08:00
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
2009-06-26 22:28:00 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
rb_link_node(&chain->rb_node, parent, p);
|
|
|
|
rb_insert_color(&chain->rb_node, root);
|
|
|
|
}
|
|
|
|
|
2009-07-05 13:39:21 +08:00
|
|
|
static void
|
|
|
|
__sort_chain_flat(struct rb_root *rb_root, struct callchain_node *node,
|
|
|
|
u64 min_hit)
|
|
|
|
{
|
|
|
|
struct callchain_node *child;
|
|
|
|
|
|
|
|
chain_for_each_child(child, node)
|
|
|
|
__sort_chain_flat(rb_root, child, min_hit);
|
|
|
|
|
|
|
|
if (node->hit && node->hit >= min_hit)
|
|
|
|
rb_insert_callchain(rb_root, node, CHAIN_FLAT);
|
|
|
|
}
|
|
|
|
|
2009-06-26 22:28:00 +08:00
|
|
|
/*
|
|
|
|
* Once we get every callchains from the stream, we can now
|
|
|
|
* sort them by hit
|
|
|
|
*/
|
2009-07-05 13:39:21 +08:00
|
|
|
static void
|
2010-08-23 02:05:22 +08:00
|
|
|
sort_chain_flat(struct rb_root *rb_root, struct callchain_root *root,
|
2009-07-05 13:39:21 +08:00
|
|
|
u64 min_hit, struct callchain_param *param __used)
|
|
|
|
{
|
2010-08-23 02:05:22 +08:00
|
|
|
__sort_chain_flat(rb_root, &root->node, min_hit);
|
2009-07-05 13:39:21 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static void __sort_chain_graph_abs(struct callchain_node *node,
|
|
|
|
u64 min_hit)
|
2009-06-26 22:28:00 +08:00
|
|
|
{
|
|
|
|
struct callchain_node *child;
|
|
|
|
|
2009-07-05 13:39:21 +08:00
|
|
|
node->rb_root = RB_ROOT;
|
2009-06-26 22:28:00 +08:00
|
|
|
|
2009-07-05 13:39:21 +08:00
|
|
|
chain_for_each_child(child, node) {
|
|
|
|
__sort_chain_graph_abs(child, min_hit);
|
2011-01-14 11:51:59 +08:00
|
|
|
if (callchain_cumul_hits(child) >= min_hit)
|
2009-07-05 13:39:21 +08:00
|
|
|
rb_insert_callchain(&node->rb_root, child,
|
|
|
|
CHAIN_GRAPH_ABS);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2010-08-23 02:05:22 +08:00
|
|
|
sort_chain_graph_abs(struct rb_root *rb_root, struct callchain_root *chain_root,
|
2009-07-05 13:39:21 +08:00
|
|
|
u64 min_hit, struct callchain_param *param __used)
|
|
|
|
{
|
2010-08-23 02:05:22 +08:00
|
|
|
__sort_chain_graph_abs(&chain_root->node, min_hit);
|
|
|
|
rb_root->rb_node = chain_root->node.rb_root.rb_node;
|
2009-07-02 23:58:21 +08:00
|
|
|
}
|
|
|
|
|
2009-07-05 13:39:21 +08:00
|
|
|
static void __sort_chain_graph_rel(struct callchain_node *node,
|
|
|
|
double min_percent)
|
2009-07-02 23:58:21 +08:00
|
|
|
{
|
|
|
|
struct callchain_node *child;
|
2009-07-05 13:39:21 +08:00
|
|
|
u64 min_hit;
|
2009-07-02 23:58:21 +08:00
|
|
|
|
|
|
|
node->rb_root = RB_ROOT;
|
2009-08-09 10:19:15 +08:00
|
|
|
min_hit = ceil(node->children_hit * min_percent);
|
2009-07-02 23:58:21 +08:00
|
|
|
|
|
|
|
chain_for_each_child(child, node) {
|
2009-07-05 13:39:21 +08:00
|
|
|
__sort_chain_graph_rel(child, min_percent);
|
2011-01-14 11:51:59 +08:00
|
|
|
if (callchain_cumul_hits(child) >= min_hit)
|
2009-07-05 13:39:21 +08:00
|
|
|
rb_insert_callchain(&node->rb_root, child,
|
|
|
|
CHAIN_GRAPH_REL);
|
2009-07-02 23:58:21 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-07-05 13:39:21 +08:00
|
|
|
static void
|
2010-08-23 02:05:22 +08:00
|
|
|
sort_chain_graph_rel(struct rb_root *rb_root, struct callchain_root *chain_root,
|
2009-07-05 13:39:21 +08:00
|
|
|
u64 min_hit __used, struct callchain_param *param)
|
2009-07-02 23:58:21 +08:00
|
|
|
{
|
2010-08-23 02:05:22 +08:00
|
|
|
__sort_chain_graph_rel(&chain_root->node, param->min_percent / 100.0);
|
|
|
|
rb_root->rb_node = chain_root->node.rb_root.rb_node;
|
2009-06-26 22:28:00 +08:00
|
|
|
}
|
|
|
|
|
2009-07-05 13:39:21 +08:00
|
|
|
int register_callchain_param(struct callchain_param *param)
|
|
|
|
{
|
|
|
|
switch (param->mode) {
|
|
|
|
case CHAIN_GRAPH_ABS:
|
|
|
|
param->sort = sort_chain_graph_abs;
|
|
|
|
break;
|
|
|
|
case CHAIN_GRAPH_REL:
|
|
|
|
param->sort = sort_chain_graph_rel;
|
|
|
|
break;
|
|
|
|
case CHAIN_FLAT:
|
|
|
|
param->sort = sort_chain_flat;
|
|
|
|
break;
|
2009-08-15 18:26:57 +08:00
|
|
|
case CHAIN_NONE:
|
2009-07-05 13:39:21 +08:00
|
|
|
default:
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-07-01 11:35:15 +08:00
|
|
|
/*
|
|
|
|
* Create a child for a parent. If inherit_children, then the new child
|
|
|
|
* will become the new parent of it's parent children
|
|
|
|
*/
|
|
|
|
static struct callchain_node *
|
|
|
|
create_child(struct callchain_node *parent, bool inherit_children)
|
2009-06-26 22:28:00 +08:00
|
|
|
{
|
|
|
|
struct callchain_node *new;
|
|
|
|
|
2010-05-10 21:56:50 +08:00
|
|
|
new = zalloc(sizeof(*new));
|
2009-06-26 22:28:00 +08:00
|
|
|
if (!new) {
|
|
|
|
perror("not enough memory to create child for code path tree");
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
new->parent = parent;
|
|
|
|
INIT_LIST_HEAD(&new->children);
|
|
|
|
INIT_LIST_HEAD(&new->val);
|
2009-07-01 11:35:15 +08:00
|
|
|
|
|
|
|
if (inherit_children) {
|
|
|
|
struct callchain_node *next;
|
|
|
|
|
|
|
|
list_splice(&parent->children, &new->children);
|
|
|
|
INIT_LIST_HEAD(&parent->children);
|
|
|
|
|
2009-07-02 23:58:19 +08:00
|
|
|
chain_for_each_child(next, new)
|
2009-07-01 11:35:15 +08:00
|
|
|
next->parent = new;
|
|
|
|
}
|
2009-06-26 22:28:00 +08:00
|
|
|
list_add_tail(&new->brothers, &parent->children);
|
|
|
|
|
|
|
|
return new;
|
|
|
|
}
|
|
|
|
|
2010-03-23 00:09:33 +08:00
|
|
|
|
2009-07-01 11:35:15 +08:00
|
|
|
/*
|
|
|
|
* Fill the node with callchain values
|
|
|
|
*/
|
2009-06-26 22:28:00 +08:00
|
|
|
static void
|
2011-01-14 11:51:58 +08:00
|
|
|
fill_node(struct callchain_node *node, struct callchain_cursor *cursor)
|
2009-06-26 22:28:00 +08:00
|
|
|
{
|
2011-01-14 11:51:58 +08:00
|
|
|
struct callchain_cursor_node *cursor_node;
|
|
|
|
|
|
|
|
node->val_nr = cursor->nr - cursor->pos;
|
|
|
|
if (!node->val_nr)
|
|
|
|
pr_warning("Warning: empty node in callchain tree\n");
|
2009-06-26 22:28:00 +08:00
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
cursor_node = callchain_cursor_current(cursor);
|
|
|
|
|
|
|
|
while (cursor_node) {
|
2009-06-26 22:28:00 +08:00
|
|
|
struct callchain_list *call;
|
|
|
|
|
2010-05-10 21:56:50 +08:00
|
|
|
call = zalloc(sizeof(*call));
|
2009-06-26 22:28:00 +08:00
|
|
|
if (!call) {
|
|
|
|
perror("not enough memory for the code path tree");
|
|
|
|
return;
|
|
|
|
}
|
2011-01-14 11:51:58 +08:00
|
|
|
call->ip = cursor_node->ip;
|
|
|
|
call->ms.sym = cursor_node->sym;
|
|
|
|
call->ms.map = cursor_node->map;
|
2009-06-26 22:28:00 +08:00
|
|
|
list_add_tail(&call->list, &node->val);
|
2011-01-14 11:51:58 +08:00
|
|
|
|
|
|
|
callchain_cursor_advance(cursor);
|
|
|
|
cursor_node = callchain_cursor_current(cursor);
|
2009-06-26 22:28:00 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2009-07-01 11:35:15 +08:00
|
|
|
static void
|
2011-01-14 11:51:58 +08:00
|
|
|
add_child(struct callchain_node *parent,
|
|
|
|
struct callchain_cursor *cursor,
|
|
|
|
u64 period)
|
2009-06-26 22:28:00 +08:00
|
|
|
{
|
|
|
|
struct callchain_node *new;
|
|
|
|
|
2009-07-01 11:35:15 +08:00
|
|
|
new = create_child(parent, false);
|
2011-01-14 11:51:58 +08:00
|
|
|
fill_node(new, cursor);
|
2009-06-26 22:28:00 +08:00
|
|
|
|
2009-08-07 13:11:05 +08:00
|
|
|
new->children_hit = 0;
|
2010-07-08 09:41:46 +08:00
|
|
|
new->hit = period;
|
2009-06-26 22:28:00 +08:00
|
|
|
}
|
|
|
|
|
2009-07-01 11:35:15 +08:00
|
|
|
/*
|
|
|
|
* Split the parent in two parts (a new child is created) and
|
|
|
|
* give a part of its callchain to the created child.
|
|
|
|
* Then create another child to host the given callchain of new branch
|
|
|
|
*/
|
2009-06-26 22:28:00 +08:00
|
|
|
static void
|
2011-01-14 11:51:58 +08:00
|
|
|
split_add_child(struct callchain_node *parent,
|
|
|
|
struct callchain_cursor *cursor,
|
|
|
|
struct callchain_list *to_split,
|
|
|
|
u64 idx_parents, u64 idx_local, u64 period)
|
2009-06-26 22:28:00 +08:00
|
|
|
{
|
|
|
|
struct callchain_node *new;
|
2009-07-01 11:35:15 +08:00
|
|
|
struct list_head *old_tail;
|
2009-07-01 18:37:06 +08:00
|
|
|
unsigned int idx_total = idx_parents + idx_local;
|
2009-06-26 22:28:00 +08:00
|
|
|
|
|
|
|
/* split */
|
2009-07-01 11:35:15 +08:00
|
|
|
new = create_child(parent, true);
|
|
|
|
|
|
|
|
/* split the callchain and move a part to the new child */
|
|
|
|
old_tail = parent->val.prev;
|
|
|
|
list_del_range(&to_split->list, old_tail);
|
|
|
|
new->val.next = &to_split->list;
|
|
|
|
new->val.prev = old_tail;
|
|
|
|
to_split->list.prev = &new->val;
|
|
|
|
old_tail->next = &new->val;
|
2009-06-26 22:28:00 +08:00
|
|
|
|
2009-07-01 11:35:15 +08:00
|
|
|
/* split the hits */
|
|
|
|
new->hit = parent->hit;
|
2009-08-07 13:11:05 +08:00
|
|
|
new->children_hit = parent->children_hit;
|
2011-01-14 11:51:59 +08:00
|
|
|
parent->children_hit = callchain_cumul_hits(new);
|
2009-07-01 11:35:15 +08:00
|
|
|
new->val_nr = parent->val_nr - idx_local;
|
|
|
|
parent->val_nr = idx_local;
|
|
|
|
|
|
|
|
/* create a new child for the new branch if any */
|
2011-01-14 11:51:58 +08:00
|
|
|
if (idx_total < cursor->nr) {
|
2009-07-01 11:35:15 +08:00
|
|
|
parent->hit = 0;
|
2011-01-14 11:51:58 +08:00
|
|
|
add_child(parent, cursor, period);
|
2010-07-08 09:41:46 +08:00
|
|
|
parent->children_hit += period;
|
2009-07-01 11:35:15 +08:00
|
|
|
} else {
|
2010-07-08 09:41:46 +08:00
|
|
|
parent->hit = period;
|
2009-07-01 11:35:15 +08:00
|
|
|
}
|
2009-06-26 22:28:00 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
2011-01-14 11:51:58 +08:00
|
|
|
append_chain(struct callchain_node *root,
|
|
|
|
struct callchain_cursor *cursor,
|
|
|
|
u64 period);
|
2009-06-26 22:28:00 +08:00
|
|
|
|
2009-07-01 11:35:15 +08:00
|
|
|
static void
|
2011-01-14 11:51:58 +08:00
|
|
|
append_chain_children(struct callchain_node *root,
|
|
|
|
struct callchain_cursor *cursor,
|
|
|
|
u64 period)
|
2009-06-26 22:28:00 +08:00
|
|
|
{
|
|
|
|
struct callchain_node *rnode;
|
|
|
|
|
|
|
|
/* lookup in childrens */
|
2009-07-02 23:58:19 +08:00
|
|
|
chain_for_each_child(rnode, root) {
|
2011-01-14 11:51:58 +08:00
|
|
|
unsigned int ret = append_chain(rnode, cursor, period);
|
2009-07-01 18:37:06 +08:00
|
|
|
|
2009-06-26 22:28:00 +08:00
|
|
|
if (!ret)
|
2009-08-07 13:11:05 +08:00
|
|
|
goto inc_children_hit;
|
2009-06-26 22:28:00 +08:00
|
|
|
}
|
2009-07-01 11:35:15 +08:00
|
|
|
/* nothing in children, add to the current node */
|
2011-01-14 11:51:58 +08:00
|
|
|
add_child(root, cursor, period);
|
2009-07-05 13:39:20 +08:00
|
|
|
|
2009-08-07 13:11:05 +08:00
|
|
|
inc_children_hit:
|
2010-07-08 09:41:46 +08:00
|
|
|
root->children_hit += period;
|
2009-06-26 22:28:00 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
2011-01-14 11:51:58 +08:00
|
|
|
append_chain(struct callchain_node *root,
|
|
|
|
struct callchain_cursor *cursor,
|
|
|
|
u64 period)
|
2009-06-26 22:28:00 +08:00
|
|
|
{
|
2011-01-14 11:51:58 +08:00
|
|
|
struct callchain_cursor_node *curr_snap = cursor->curr;
|
2009-06-26 22:28:00 +08:00
|
|
|
struct callchain_list *cnode;
|
2011-01-14 11:51:58 +08:00
|
|
|
u64 start = cursor->pos;
|
2009-06-26 22:28:00 +08:00
|
|
|
bool found = false;
|
2011-01-14 11:51:58 +08:00
|
|
|
u64 matches;
|
2009-06-26 22:28:00 +08:00
|
|
|
|
2009-07-01 11:35:15 +08:00
|
|
|
/*
|
|
|
|
* Lookup in the current node
|
|
|
|
* If we have a symbol, then compare the start to match
|
|
|
|
* anywhere inside a function.
|
|
|
|
*/
|
2009-06-26 22:28:00 +08:00
|
|
|
list_for_each_entry(cnode, &root->val, list) {
|
2011-01-14 11:51:58 +08:00
|
|
|
struct callchain_cursor_node *node;
|
2010-03-23 00:09:33 +08:00
|
|
|
struct symbol *sym;
|
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
node = callchain_cursor_current(cursor);
|
|
|
|
if (!node)
|
2009-07-01 11:35:15 +08:00
|
|
|
break;
|
2010-03-23 00:09:33 +08:00
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
sym = node->sym;
|
2010-03-23 00:09:33 +08:00
|
|
|
|
2010-03-25 03:40:18 +08:00
|
|
|
if (cnode->ms.sym && sym) {
|
|
|
|
if (cnode->ms.sym->start != sym->start)
|
2009-07-01 11:35:15 +08:00
|
|
|
break;
|
2011-01-14 11:51:58 +08:00
|
|
|
} else if (cnode->ip != node->ip)
|
2009-06-26 22:28:00 +08:00
|
|
|
break;
|
2010-03-23 00:09:33 +08:00
|
|
|
|
2009-06-26 22:28:00 +08:00
|
|
|
if (!found)
|
|
|
|
found = true;
|
2011-01-14 11:51:58 +08:00
|
|
|
|
|
|
|
callchain_cursor_advance(cursor);
|
2009-06-26 22:28:00 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
/* matches not, relay on the parent */
|
2011-01-14 11:51:58 +08:00
|
|
|
if (!found) {
|
|
|
|
cursor->curr = curr_snap;
|
|
|
|
cursor->pos = start;
|
2009-06-26 22:28:00 +08:00
|
|
|
return -1;
|
2011-01-14 11:51:58 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
matches = cursor->pos - start;
|
2009-06-26 22:28:00 +08:00
|
|
|
|
|
|
|
/* we match only a part of the node. Split it and add the new chain */
|
2011-01-14 11:51:58 +08:00
|
|
|
if (matches < root->val_nr) {
|
|
|
|
split_add_child(root, cursor, cnode, start, matches, period);
|
2009-06-26 22:28:00 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* we match 100% of the path, increment the hit */
|
2011-01-14 11:51:58 +08:00
|
|
|
if (matches == root->val_nr && cursor->pos == cursor->nr) {
|
2010-07-08 09:41:46 +08:00
|
|
|
root->hit += period;
|
2009-06-26 22:28:00 +08:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2009-07-01 11:35:15 +08:00
|
|
|
/* We match the node and still have a part remaining */
|
2011-01-14 11:51:58 +08:00
|
|
|
append_chain_children(root, cursor, period);
|
2009-07-01 11:35:15 +08:00
|
|
|
|
|
|
|
return 0;
|
2009-06-26 22:28:00 +08:00
|
|
|
}
|
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
int callchain_append(struct callchain_root *root,
|
|
|
|
struct callchain_cursor *cursor,
|
|
|
|
u64 period)
|
2009-06-26 22:28:00 +08:00
|
|
|
{
|
2011-01-14 11:51:58 +08:00
|
|
|
if (!cursor->nr)
|
2010-03-23 00:09:33 +08:00
|
|
|
return 0;
|
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
callchain_cursor_commit(cursor);
|
2010-03-23 00:09:33 +08:00
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
append_chain_children(&root->node, cursor, period);
|
2010-08-23 02:05:22 +08:00
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
if (cursor->nr > root->max_depth)
|
|
|
|
root->max_depth = cursor->nr;
|
2010-03-23 00:09:33 +08:00
|
|
|
|
|
|
|
return 0;
|
2009-06-26 22:28:00 +08:00
|
|
|
}
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
|
|
|
|
static int
|
2011-01-14 11:51:58 +08:00
|
|
|
merge_chain_branch(struct callchain_cursor *cursor,
|
|
|
|
struct callchain_node *dst, struct callchain_node *src)
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
{
|
2011-01-14 11:51:58 +08:00
|
|
|
struct callchain_cursor_node **old_last = cursor->last;
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
struct callchain_node *child, *next_child;
|
|
|
|
struct callchain_list *list, *next_list;
|
2011-01-14 11:51:58 +08:00
|
|
|
int old_pos = cursor->nr;
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
int err = 0;
|
|
|
|
|
|
|
|
list_for_each_entry_safe(list, next_list, &src->val, list) {
|
2011-01-14 11:51:58 +08:00
|
|
|
callchain_cursor_append(cursor, list->ip,
|
|
|
|
list->ms.map, list->ms.sym);
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
list_del(&list->list);
|
|
|
|
free(list);
|
|
|
|
}
|
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
if (src->hit) {
|
|
|
|
callchain_cursor_commit(cursor);
|
|
|
|
append_chain_children(dst, cursor, src->hit);
|
|
|
|
}
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
|
|
|
|
chain_for_each_child_safe(child, next_child, src) {
|
2011-01-14 11:51:58 +08:00
|
|
|
err = merge_chain_branch(cursor, dst, child);
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
if (err)
|
|
|
|
break;
|
|
|
|
|
|
|
|
list_del(&child->brothers);
|
|
|
|
free(child);
|
|
|
|
}
|
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
cursor->nr = old_pos;
|
|
|
|
cursor->last = old_last;
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
int callchain_merge(struct callchain_cursor *cursor,
|
|
|
|
struct callchain_root *dst, struct callchain_root *src)
|
|
|
|
{
|
|
|
|
return merge_chain_branch(cursor, &dst->node, &src->node);
|
|
|
|
}
|
|
|
|
|
|
|
|
int callchain_cursor_append(struct callchain_cursor *cursor,
|
|
|
|
u64 ip, struct map *map, struct symbol *sym)
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
{
|
2011-01-14 11:51:58 +08:00
|
|
|
struct callchain_cursor_node *node = *cursor->last;
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
if (!node) {
|
|
|
|
node = calloc(sizeof(*node), 1);
|
|
|
|
if (!node)
|
|
|
|
return -ENOMEM;
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
*cursor->last = node;
|
|
|
|
}
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
node->ip = ip;
|
|
|
|
node->map = map;
|
|
|
|
node->sym = sym;
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
cursor->nr++;
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
|
2011-01-14 11:51:58 +08:00
|
|
|
cursor->last = &node->next;
|
|
|
|
|
|
|
|
return 0;
|
perf: Support for callchains merge
If we sort the histograms by comm, which is the default,
we need to merge some of them, typically different thread
histograms of a same process, or just same comm. But during
this merge, we forgot to merge callchains.
So imagine we have three threads (tids: 1000, 1001, 1002) that
belong to comm "foo".
tid 1000 got 100 events
tid 1001 got 10 events
tid 1002 got 3 events
Once we merge these histograms to get a per comm result, we'll
finally get:
"foo" got 113 events
The problem is if we merge 1000 and 1001 histograms into 1002, then
the end merge result, wrt callchains, will be only callchains that
belong to 1002.
This is because we haven't handled callchains in the merge. Only those
from one of the threads inside a common comm survive.
It means during this merge, we can lose a lot of callchains.
Fix this by implementing callchains merge and apply it on histograms
that collapse.
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Paul Mackerras <paulus@samba.org>
2010-08-23 03:10:35 +08:00
|
|
|
}
|