When change it with the following:
name = NULL;
The whole gdb backtrace is:
#0 0xb804f424 in __kernel_vsyscall ()
#1 0x00b7dd80 in raise () from /lib/libc.so.6
#2 0x00b7f691 in abort () from /lib/libc.so.6
#3 0x00bb624b in __libc_message () from /lib/libc.so.6
#4 0x00bbe0f1 in _int_free () from /lib/libc.so.6
#5 0x00bc1bc0 in free () from /lib/libc.so.6
#6 0x0805d289 in perf_event::set_event_name (this=0xc94ff4,
event_name=0x8089a6b "power:power_frequency") at perf/perf.cpp:153
#7 0x0805d752 in perf_bundle::add_event (this=0x8097de0,
event_name=0x8089a6b "power:power_frequency") at perf/perf_bundle.cpp:68
#8 0x08054e7d in enumerate_cpus () at cpu/cpu.cpp:242
#9 0x0804e77d in main (argc=1, argv=0xbff2a7b4) at main.cpp:175
I think free "power:power_frequency" is the porblem.
At 2011-01-01 08:35:00，wkq5325 <wkq5...@163.com> wrote:
I get code from
git clone git://git.kernel.org/pub/scm/status/powertop/powertop.git
The whole gdb backtrace is:
#0 0xb80ad424 in __kernel_vsyscall ()
#1 0x00b7dd80 in raise () from /lib/libc.so.6
#2 0x00b7f691 in abort () from /lib/libc.so.6
#3 0x00bb624b in __libc_message () from /lib/libc.so.6
#4 0x00bbe0f1 in _int_free () from /lib/libc.so.6
#5 0x00bc1bc0 in free () from /lib/libc.so.6
#6 0x0805d289 in perf_event::set_event_name (this=0xc94ff4,
event_name=0x8089a4b "power:power_frequency") at perf/perf.cpp:152
#7 0x0805d732 in perf_bundle::add_event (this=0x8097de0,
event_name=0x8089a4b "power:power_frequency") at perf/perf_bundle.cpp:68
#8 0x08054e7d in enumerate_cpus () at cpu/cpu.cpp:242
#9 0x0804e77d in main (argc=1, argv=0xbfaadf34) at main.cpp:175
Happy new year!
At 2011-01-01 00:01:28，"Sergey Senozhatsky" <sergey.senozhat...@gmail.com>
wrote:
>On (12/31/10 18:28), wkq5325 wrote:
>> Sorry, it can't work with
>>
>> name = NULL;
>>
>
>Hm, could you please provide the whole gdb backtrace (bt)?
>
>I thought it will work. We don't have pointer initialization in .ctor
>so it could be anything and could pass `if (name)' check (which will
>cause freeing of unallocated memory).
>
>Do we have the exact stack traces for patched and unpatched versions?
>
>Happy new year!
>
>
>> Same problem like before.
>>
>> >No, changing it to `if (!name)' will introduce bug.
>> >Could you please try the following one:
>> >
>> >---
>> >
>> > perf/perf.cpp | 1 +
>> > 1 files changed, 1 insertions(+), 0 deletions(-)
>> >
>> >diff --git a/perf/perf.cpp b/perf/perf.cpp
>> >index c9758ff..2f2686b 100644
>> >--- a/perf/perf.cpp
>> >+++ b/perf/perf.cpp
>> >@@ -177,6 +177,7 @@ void perf_event::set_cpu(int _cpu)
>> >
>> > perf_event::perf_event(const char *event_name, int _cpu, int buffer_size)
>> > {
>> >+ name = NULL;
>> > set_event_name(event_name);
>> > perf_fd = -1;
>> > bufsize = buffer_size;
>> >
>
>
> Sergey
_______________________________________________
Discuss mailing list
Discuss@lesswatts.org
http://lists.lesswatts.org/listinfo/discuss
