This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: [PATCH 2.5/4 v2] GAS: Make new fake labels when cloning a symbol
- From: "Maciej W. Rozycki" <macro at codesourcery dot com>
- To: Richard Earnshaw <rearnsha at arm dot com>
- Cc: Richard Sandiford <rdsandiford at googlemail dot com>, Jie Zhang <jie at codesourcery dot com>, Catherine Moore <clm at codesourcery dot com>, binutils at sourceware dot org
- Date: Thu, 2 Dec 2010 23:18:43 +0000 (GMT)
- Subject: Re: [PATCH 2.5/4 v2] GAS: Make new fake labels when cloning a symbol
- References: <alpine.DEB.1.10.1007241721120.29495@tp.orcam.me.uk> <alpine.DEB.1.10.1010291442440.25860@tp.orcam.me.uk> <87aalwngmr.fsf@firetop.home> <alpine.DEB.1.10.1011011201550.27998@tp.orcam.me.uk> <87r5f4d92p.fsf@firetop.home> <alpine.DEB.1.10.1012012126420.14129@tp.orcam.me.uk> <1291309254.9299.9.camel@e102346-lin.cambridge.arm.com>
Hi Richard,
> > > Sounds good, but please don't let it get in the way of applying this patch
> > > (when the time comes). The change above seems entirely in keeping with
> > > the way that abs_symbol is initialised.
> >
> > This is the version I applied.
> >
> > 2010-12-01 Maciej W. Rozycki <macro@codesourcery.com>
> >
> > * symbols.h (dot_symbol): New declaration.
> > (dot_symbol_init): New prototype.
> > * symbols.c (dot_symbol): New variable.
> > (symbol_clone): Assert it's not dot_symbol being cloned.
> > (dot_symbol_init): New function.
> > (symbol_clone_if_forward_ref): Create a new temporary symbol
> > when trying to clone dot_symbol.
> > * expr.c (current_location): Refer to dot_symbol instead of
> > making a new temporary symbol.
> > * read.c (read_a_source_file): Update dot_symbol as we go.
> > * as.c (main): Call dot_symbol_init.
>
> So I'm not entirely sure that this patch is responsible, but it does
> look a likely candidate...
>
> .section .text._ZnajPv,"axG",%progbits,_ZnajPv,comdat
> .fnstart
> .fnend
>
> /arm/scratch/rearnsha/gnu/gcc-results/trunk/gcc/../gas/as-new
> -o /tmp/x.o test.s
> test.s: Assembler messages:
> test.s:3: Error: redefined symbol cannot be used on reloc
> /arm/scratch/rearnsha/gnu/gcc-results/trunk/gcc/../gas/as-new: /tmp/x.o:
> symbol `.' required but not present
> test.s:3: Fatal error: can't close /tmp/x.o: No symbols
>
> This is on arm-eabi.
Thanks for reducing the test case. Below I'm including a change that is
supposed to fix it. The original change indeed is what triggered the
problem, but from the code I have modified I infer it is more related to
the issue with equated symbols not being handled correctly that I have
fixed as well. This function counts as a symbol reference and as such
should resolve any equated symbols by making a clone. Chances are code
could have been crafted that would trigger a problem here even before my
fixes, but I found no justification to spend time investigating that.
The change below fixes the problem with your test case as well as one
attached to PR gas/12282. I have regression-tested it with the
arm-none-eabi and mips-sde-elf targets.
2010-12-02 Maciej W. Rozycki <macro@codesourcery.com>
PR gas/12282
* expr.c (make_expr_symbol): Make a clone if handling an
equated symbol.
OK to apply?
Maciej
binutils-gas-dot-fix.diff
Index: binutils-fsf-trunk-quilt/gas/expr.c
===================================================================
--- binutils-fsf-trunk-quilt.orig/gas/expr.c 2010-12-02 22:10:55.000000000 +0000
+++ binutils-fsf-trunk-quilt/gas/expr.c 2010-12-02 22:11:31.000000000 +0000
@@ -76,7 +76,7 @@ make_expr_symbol (expressionS *expressio
if (expressionP->X_op == O_symbol
&& expressionP->X_add_number == 0)
- return expressionP->X_add_symbol;
+ return symbol_clone_if_forward_ref (expressionP->X_add_symbol);
if (expressionP->X_op == O_big)
{