Re: does gcc gives a false warning in kernel/trace/trace_events_filter.c ?

["Toralf Förster" <toralf.foerster@gmx.de>]

On 09/05/2012 07:08 PM, Steven Rostedt wrote:
> On Sun, 2012-09-02 at 11:04 +0200, Toralf Förster wrote:
>> The current git tree of linux gave with gcc-4.6.3 :
>>
>> kernel/trace/trace_events_filter.c: In function ‘ftrace_function_set_filter_cb’:
>> kernel/trace/trace_events_filter.c:2074:8: warning: ‘ret’ may be used uninitialized in this function [-Wuninitialized] 
>>
>>
>> which refers to this piece of code:
>>
>>
>>   2061  static int ftrace_function_set_filter_cb(enum move_type move,
>>   2062                                           struct filter_pred *pred,
>>   2063                                           int *err, void *data)
>>   2064  {
>>   2065          /* Checking the node is valid for function trace. */
>>   2066          if ((move != MOVE_DOWN) ||
>>   2067              (pred->left != FILTER_PRED_INVALID)) {
>>   2068                  *err = ftrace_function_check_pred(pred, 0);
>>   2069          } else {
>>   2070                  *err = ftrace_function_check_pred(pred, 1);
>>   2071                  if (*err)
>>   2072                          return WALK_PRED_ABORT;
>>   2073 
>>   2074                  *err = __ftrace_function_set_filter(pred->op == OP_EQ,
>>   2075                                                      pred->regex.pattern,
>>   2076                                                      pred->regex.len,
>>   2077                                                      data);
>>   2078          }
>>   2079 
>>   2080          return (*err) ? WALK_PRED_ABORT : WALK_PRED_DEFAULT;
>>   2081  }
>>   2082  
>>
>>
>> >From a Gentoo forum user I got a hint :
>>
>> "Maybe it's some kind of a weird inlining issue? I think it's
>> referring to the ret in __ftrace_function_set_filter(), which would be
>> uninitialized if the for-loop does not run (re_cnt ≤ 0)"
>>
>> Now I'm wondering if re_cnt can become zero or if gcc is wrong here ?
>>
> 
> Strange, as ret is initialized to 'ret = -EINVAL;' in
> __ftrace_function_set_filter(). I'm thinking that gcc got confused here.
> Maybe report it to the gcc maintainers?
> 
> -- Steve
> 

I filed a bug report

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54495
and got this answer : 

--- Comment #1 from Hans-Peter Nilsson <hp at gcc dot gnu.org> 2012-09-05 22:14:00 UTC ---
But if the call to ftrace_function_filter_re sets re_cnt to 0, then ret indeed
will be used uninitialized AFAICT.  What am I missing?

-- 
MfG/Sincerely
Toralf Förster
pgp finger print: 7B1A 07F4 EC82 0F90 D4C2 8936 872A E508 7DB6 9DA3