Re: RFA: Recognize bottom of stack on Linux

[Daniel Jacobowitz <dmj+ at andrew dot cmu dot edu>]

On Wed, Feb 06, 2002 at 02:26:00PM -0500, Jim Blandy wrote:
> 
> Daniel Jacobowitz <drow@mvista.com> writes:
> > I like this.  The way func_frame_chain_valid should really be used is
> > by something like:
> > 
> >   /* NOTE: tm-i386nw.h and tm-i386v4.h override this.  */
> >   set_gdbarch_frame_chain_valid (gdbarch, file_frame_chain_valid);
> > 
> > (copied from i386-tdep.c).
> > 
> > Does this patch work for you?
> 
> Yes, thanks.  I've included a revised version of my patch below.
> 
> > I'm curious as to why we can't just set this universally, or at least a
> > little more globally.  Most things that have a main () use it as a
> > normal main ().  I'd propose that we set it as the default frame chain,
> > and provide/document an option to ignore inside_main_func.
> 
> Well, gdbarch is never supposed to change the default behavior of
> macros; this helps us convert pre-gdbarch targets incrementally.
> Simply turning on gdbarch for one's target ideally wouldn't change its
> behavior at all.

Ping - this patch seems to have fallen through the cracks.  It fixes
eight of the twenty remaining testsuite failures on my configuration,
so I'd like to see it committed.

> 
> 
> 2002-02-06  Jim Blandy  <jimb@redhat.com>
> 
> 	Fix from Daniel Jacobowitz:
> 	* config/i386/tm-linux.h (FRAME_CHAIN_VALID): Use
> 	`func_frame_chain_valid' for this.
> 	* i386-tdep.c (i386_gdbarch_init): Note that tm-linux.h is
> 	overriding the selection we make here.
> 
> Index: gdb/i386-tdep.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/i386-tdep.c,v
> retrieving revision 1.50
> diff -c -r1.50 i386-tdep.c
> *** gdb/i386-tdep.c	2002/01/01 16:29:43	1.50
> --- gdb/i386-tdep.c	2002/02/06 19:19:27
> ***************
> *** 1324,1330 ****
>   
>     set_gdbarch_pc_in_call_dummy (gdbarch, pc_in_call_dummy_on_stack);
>   
> !   /* NOTE: tm-i386nw.h and tm-i386v4.h override this.  */
>     set_gdbarch_frame_chain_valid (gdbarch, file_frame_chain_valid);
>   
>     /* NOTE: tm-i386aix.h, tm-i386bsd.h, tm-i386os9k.h, tm-linux.h,
> --- 1324,1330 ----
>   
>     set_gdbarch_pc_in_call_dummy (gdbarch, pc_in_call_dummy_on_stack);
>   
> !   /* NOTE: tm-i386nw.h, tm-i386v4.h, and tm-linux.h override this.  */
>     set_gdbarch_frame_chain_valid (gdbarch, file_frame_chain_valid);
>   
>     /* NOTE: tm-i386aix.h, tm-i386bsd.h, tm-i386os9k.h, tm-linux.h,
> Index: gdb/config/i386/tm-linux.h
> ===================================================================
> RCS file: /cvs/src/src/gdb/config/i386/tm-linux.h,v
> retrieving revision 1.16
> diff -c -r1.16 tm-linux.h
> *** gdb/config/i386/tm-linux.h	2001/11/08 00:03:52	1.16
> --- gdb/config/i386/tm-linux.h	2002/02/06 19:19:28
> ***************
> *** 82,87 ****
> --- 82,98 ----
>   #define IN_SIGTRAMP(pc, name) i386_linux_in_sigtramp (pc, name)
>   extern int i386_linux_in_sigtramp (CORE_ADDR, char *);
>   
> + /* On Linux, the entry point is called _start, but that invokes
> +    something called __libc_start_main, which calls main.  So if we
> +    want the stack to end at main (as it does for most GDB targets),
> +    it's not enough for us to use inside_entry_func or
> +    inside_entry_file; that rule will only trigger after we've included
> +    __libc_start_main in the backtrace, which we don't want.
> +    func_frame_chain_valid checks both for `main', and for the entry
> +    point function.  */
> + /* Use the alternate method of determining valid frame chains. */
> + #define FRAME_CHAIN_VALID(fp,fi) func_frame_chain_valid (fp, fi)
> + 
>   #undef FRAME_CHAIN
>   #define FRAME_CHAIN(frame) i386_linux_frame_chain (frame)
>   extern CORE_ADDR i386_linux_frame_chain (struct frame_info *frame);
> 

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer