Hi Jim,
Thanks for the response. Indeed, there's no section change that would
position label LFE17 in the .text section, where main is.
On Sat, 2007-04-21 at 10:35 -0700, Jim Wilson wrote:
> It looks like you are trying to do some complicated (and probably
> fragile) section switching stuff in a C function using gcc extended
> asms, and that there is an unbalanced section switch somewhere. Or
> maybe gcc optimization is breaking your code.
I wonder if this is a result of dead-code elimination by gcc, and if
there's a way to change this behavior. I've traced this down to the
following C code:
==============================================
hcu_wrapped_main_0_before:
/* more stuff */
if (var == 1)
{
hcu_save_frame_of_main_v1 ((void *) (&hcu_stack_local_main_var),
&hcu_stack_local_main_var.
hcu_stack_frame_fields, 2, 7);
return;
hcu_wrapped_main_2_after:;
}
goto hcu_wrapped_main_0_before;
__asm__ __volatile__ ("hcu_uncommon_section_"
"main_v1_hcu_wrapped_main_0_call_must_not"
":\n\t");;
__asm__ __volatile__ ("jmp 1f\n\t" ".text\n\t" "1:\n");;
==============================================
It seems that gcc does not even produce the inline assembly after the
last goto, perhaps because it is deemed that it's dead code. However, I
need this assembly produced to emit the section change to ".text". Is
there a way to tell gcc to NOT remove that inline assembly ? Is there
some attribute/keyword that describes this ? I haven't found one.
The only other alternative I see is to acquire the value of the
hcu_wrapped_main_0_before label using the && operator
(http://gcc.gnu.org/onlinedocs/gcc/Labels-as-Values.html) and write
something like:
__asm__ __volatile__ ("jmp %0\n\t"
"hcu_uncommon_section_"
"main_v1_hcu_wrapped_main_0_call_must_not"
":\n\t"
: /* no output */
: "g" (&&hcu_wrapped_main_0_before) );;
__asm__ __volatile__ ("jmp 1f\n\t" ".text\n\t" "1:\n");;
It's not an ideal solution for two reasons. Labels as values are not ISO
C. And these assembly statements are produced using macros that could be
conditionally defined to something a lot more simpler (just a "}") when
cache locality is not an issue. I was hoping to keep those macros.
Thanks again,
Kristis
_______________________________________________
bug-binutils mailing list
bug-binutils@gnu.org
http://lists.gnu.org/mailman/listinfo/bug-binutils
