perf tools: callchain: Fix bad rounding of minimum rate
Frederic Weisbecker [Sun, 9 Aug 2009 02:19:15 +0000 (04:19 +0200)]
Sometimes we get callchain branches that have a rate under the
limit given by the user.

Say you launched:

 perf record -f -g -a ./hackbench 10
 perf report -g fractal,10.0

And you got:

2.33%       hackbench  [kernel]                  [k] _spin_lock_irqsave
                |
                |--78.57%-- remove_wait_queue
                |          poll_freewait
                |          do_sys_poll
                |          sys_poll
                |          sysenter_dispatch
                |          0xf7ffa430
                |          0x1ffadea3c
                |
                |--7.14%-- __up_read
                |          up_read
                |          do_page_fault
                |          page_fault
                |          0xf7ffa430
                |          0xa0df710000000a
                ...

It is abnormal to get a 7.14% branch whereas we passed a 10%
filter.

The problem is that we round down the minimum threshold. This
happens mostly when we have very low number of events. If the
total amount of your branch is 4 and you have a subranch of 3
events, filtering to 90% will be computed like follows:

  limit = 4 * 0.9;

The result is about 3.6, but the cast to integer will round
down to 3. It means that our filter is actually of 75%

We must then explicitly round up the minimum threshold.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: acme@redhat.com
Cc: peterz@infradead.org
Cc: efault@gmx.de
LKML-Reference: <20090809024235.GA10146@nowhere>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

tools/perf/util/callchain.c

index a8e67aa..0114734 100644 (file)
@@ -13,6 +13,7 @@
 #include <stdio.h>
 #include <stdbool.h>
 #include <errno.h>
+#include <math.h>
 
 #include "callchain.h"
 
@@ -112,7 +113,7 @@ static void __sort_chain_graph_rel(struct callchain_node *node,
        u64 min_hit;
 
        node->rb_root = RB_ROOT;
-       min_hit = node->children_hit * min_percent / 100.0;
+       min_hit = ceil(node->children_hit * min_percent);
 
        chain_for_each_child(child, node) {
                __sort_chain_graph_rel(child, min_percent);
@@ -126,7 +127,7 @@ static void
 sort_chain_graph_rel(struct rb_root *rb_root, struct callchain_node *chain_root,
                     u64 min_hit __used, struct callchain_param *param)
 {
-       __sort_chain_graph_rel(chain_root, param->min_percent);
+       __sort_chain_graph_rel(chain_root, param->min_percent / 100.0);
        rb_root->rb_node = chain_root->rb_root.rb_node;
 }