Home Original page

building glibc with gold linker & -frecord-gcc-switches results in internal error

Cary Coutant ccoutant@google.com
Tue Feb 14 23:30:00 GMT 2012 
> The Makefile rules for crti.o do some massaging of the assembly with
> sed. I wonder if something is going haywire there.

That's exactly what's happening. Here's the code for initfini.s...

	.file	"initfini.c"
	.section	.GCC.command.line,"MS",@progbits,1
	.ascii	"-nostdinc"
	.zero	1
	.ascii	"-I ../include"
	.zero	1
	.ascii	"-I /var/tmp/portage/cross-i686-pc-linux-gnu/glibc-2.11.1-r3/"
	.ascii	"work/build-default-i686-pc-linux-gnu-nptl/csu"
	.zero	1
	[...snip...]

	.ascii	"-finhibit-size-directive"
	.zero	1
	.ascii	"-fno-exceptions"
	.zero	1
#APP
	
#include "defs.h"
	
/*@HEADER_ENDS*/
	
/*@TESTS_BEGIN*/
#NO_APP
	.text
	.p2align 2,,3
	.globl	dummy
	.type	dummy, @function
dummy:
	pushl	%ebp
	movl	%esp, %ebp
	subl	$8, %esp
	movl	8(%ebp), %eax
	testl	%eax, %eax
	je	.L1
	call	*%eax
.L1:
	leave
	ret
#APP
	
/*@TESTS_END*/
	
/*@_init_PROLOG_BEGINS*/
#NO_APP
	.p2align 2,,3
	.type	call_gmon_start, @function
call_gmon_start:
	pushl	%ebp
	movl	%esp, %ebp
	pushl	%ebx
	pushl	%edx
	call	.L6
	[...]

The glibc/csu/Makefile runs this code through a sed script that
deletes everything between HEADER_ENDS and _init_PROLOG_BEGINS, which
loses the ".text" directive, and the code for call_gmon_start ends up
placed in the .GCC.command.line section.

Refile this as a glibc bug (although we'll keep this one open for the
assert failure that should be a reasonable error message).

-cary
More information about the Binutils mailing list