Re: Backtrace of naked functions

[Ruppert <dieter_ruppert at siemens dot com>]

>
>> Because we use the stabs debugging format, we could simply link in
>> a module which defines a 'main stab'. gdb honors this stab type, and
>> treats the name given with this stab like 'main'. In particular,
>> backtraces stop at this function.
>
>Are you using any particular compiler/linker switches to achieve this?
>I can only get this to work if I patch GDB's dbxread.c.

Ah yes. The N_MAIN stab is seen in process_one_symbol (in dbxread.c), and
the name given ends up in the variable name_of_main. I remember vaguely
that this worked only with gdb --readnow; to have this available
in the general case we needed also a patch like the one you proposed.
Sorry, I completely forgot this.

>
>> This may be restricted to the stabs debugging format, though, and it
>> won't help if you don't have such a "topmost" function.
>
>And it won't help if you have multiple such functions. Otherwise the 
>trick is neat, even though a work-around for the initial problem.
>

Yes. This is definitely only a workaround. 

I also investigated this for a while, and I came up with the following
hack, which works when the stack is terminated with two zero
values (for previous ebp and return address):

 - at the very beginning of function inside_main_func (in blockframe.c)
   insert
   
    if ((pc == 0) || (pc == 0xffffffff))
       return 1; 
       
 - at the end of get_prev_frame (in frame.c), replace
 
  /* Link it in.  */
  this_frame->prev = prev_frame;
  prev_frame->next = this_frame;

  by
  
  /* Link it in.  */
  prev_frame->next = this_frame;
  if (get_frame_pc(prev_frame) == 0)
     return NULL;
  this_frame->prev = prev_frame;

This hack may or may not work in your case (we use gdb-6.1), so use with
caution.

Regards
D.Ruppert
ePS&RTS GmbH
dieter_ruppert@siemens.com



------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sources.redhat.com