See http://sources.redhat.com/ml/libc-hacker/2005-03/msg00008.html From libc-hacker-return-8343-listarch-libc-hacker=sources dot redhat dot com at sources dot redhat dot com Sat Mar 05 09:21:18 2005 Return-Path: Delivered-To: listarch-libc-hacker at sources dot redhat dot com Received: (qmail 23370 invoked by alias); 5 Mar 2005 09:21:15 -0000 Mailing-List: contact libc-hacker-help at sources dot redhat dot com; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: libc-hacker-owner at sources dot redhat dot com Delivered-To: mailing list libc-hacker at sources dot redhat dot com Received: (qmail 22971 invoked from network); 5 Mar 2005 09:20:51 -0000 Received: from unknown (HELO sunsite.mff.cuni.cz) (195.113.15.26) by sourceware dot org with SMTP; 5 Mar 2005 09:20:51 -0000 Received: from sunsite.mff.cuni.cz (sunsite.mff.cuni.cz [127.0.0.1]) by sunsite dot mff dot cuni dot cz (8 dot 13 dot 1/8 dot 13 dot 1) with ESMTP id j259KoB5020894; Sat, 5 Mar 2005 10:20:50 +0100 Received: (from jj@localhost) by sunsite dot mff dot cuni dot cz (8 dot 13 dot 1/8 dot 13 dot 1/Submit) id j259KnO1020889; Sat, 5 Mar 2005 10:20:49 +0100 Date: Sat, 5 Mar 2005 10:20:49 +0100 From: Jakub Jelinek To: Ulrich Drepper , Roland McGrath Cc: Glibc hackers Subject: [PATCH] Fix build with GCC 4 Message-ID: <20050305092049.GJ4777@sunsite.mff.cuni.cz> Reply-To: Jakub Jelinek Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.4.1i Hi! The thread_offsetof change just mirrors what Alan Modra did to NPTL tcb-offsets.sym. Apparently for GCC 4 an offsetof like expression, but not really offsetof, is no longer constant folded and therefore not suitable for "i" constraint. The ELF_MACHINE_NO_RELA change is needed to avoid rtld.c: In function '_dl_start': dynamic-link.h:50: error: nested function 'elf_machine_rela_relative' declared but never defined dynamic-link.h:47: error: nested function 'elf_machine_rela' declared but never defined This is what happens. rtld.c first includes dl-machine.h without RESOLVE_MAP and without RTLD_BOOTSTRAP defined. This means that ELF_MACHINE_NO_RELA is not defined on i386/arm. Later on it defines RESOLVE_MAP and RTLD_BOOTSTRAP and includes dynamic-link.h which has: # if ! ELF_MACHINE_NO_RELA auto void __attribute__((always_inline)) elf_machine_rela (struct link_map *map, const ElfW(Rela) *reloc, const ElfW(Sym) *sym, const struct r_found_version *version, void *const reloc_addr); auto void __attribute__((always_inline)) elf_machine_rela_relative (ElfW(Addr) l_addr, const ElfW(Rela) *reloc, void *const reloc_addr); # endif and later on includes dl-machine.h which sees RTLD_BOOTSTRAP is defined and defines ELF_MACHINE_NO_RELA and doesn't define elf_machine_rela* nested functions. But the prototypes were already defined and GCC 4 doesn't like this. ELF_MACHINE_NO_RELA is only ever used in preprocessing conditionals and never in defined ELF_MACHINE_NO_RELA, so the trick below already defines ELF_MACHINE_NO_RELA to 1/0 depending on whether RTLD_BOOTSTRAP is defined and thus the prototypes in dynamic-link.h that are not desirable are gone. 2005-03-05 Jakub Jelinek * sysdeps/i386/dl-machine.h (ELF_MACHINE_NO_RELA): Define unconditionally to (defined RTLD_BOOTSTRAP). * sysdeps/arm/dl-machine.h (ELF_MACHINE_NO_RELA): Likewise. linuxthreads/ * sysdeps/powerpc/tcb-offsets.sym (thread_offsetof): Rework for GCC 4. --- libc/linuxthreads/sysdeps/powerpc/tcb-offsets.sym.jj 2005-03-04 14:21:29.000000000 -0500 +++ libc/linuxthreads/sysdeps/powerpc/tcb-offsets.sym 2005-03-04 14:26:29.000000000 -0500 @@ -8,7 +8,7 @@ -- Abuse tls.h macros to derive offsets relative to the thread register. # undef __thread_register # define __thread_register ((void *) 0) -# define thread_offsetof(mem) ((void *) &THREAD_SELF->p_##mem - (void *) 0) +# define thread_offsetof(mem) ((ptrdiff_t) THREAD_SELF + offsetof (struct _pthread_descr_struct, p_##mem)) # else