From mboxrd@z Thu Jan 1 00:00:00 1970 X-Msuck: nntp://news.gmane.org/gmane.linux.lib.musl.general/7452 Path: news.gmane.org!not-for-mail From: Rich Felker Newsgroups: gmane.linux.lib.musl.general Subject: Re: x86_64 and x32 fail to build Date: Mon, 20 Apr 2015 17:13:59 -0400 Message-ID: <20150420211359.GS6817@brightrain.aerifal.cx> References: Reply-To: musl@lists.openwall.com NNTP-Posting-Host: plane.gmane.org Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Trace: ger.gmane.org 1429564457 24010 80.91.229.3 (20 Apr 2015 21:14:17 GMT) X-Complaints-To: usenet@ger.gmane.org NNTP-Posting-Date: Mon, 20 Apr 2015 21:14:17 +0000 (UTC) To: musl@lists.openwall.com Original-X-From: musl-return-7465-gllmg-musl=m.gmane.org@lists.openwall.com Mon Apr 20 23:14:17 2015 Return-path: Envelope-to: gllmg-musl@m.gmane.org Original-Received: from mother.openwall.net ([195.42.179.200]) by plane.gmane.org with smtp (Exim 4.69) (envelope-from ) id 1YkJ1D-0006iA-G8 for gllmg-musl@m.gmane.org; Mon, 20 Apr 2015 23:14:15 +0200 Original-Received: (qmail 24059 invoked by uid 550); 20 Apr 2015 21:14:14 -0000 Mailing-List: contact musl-help@lists.openwall.com; run by ezmlm Precedence: bulk List-Post: List-Help: List-Unsubscribe: List-Subscribe: Original-Received: (qmail 24041 invoked from network); 20 Apr 2015 21:14:13 -0000 Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.21 (2010-09-15) Original-Sender: Rich Felker Xref: news.gmane.org gmane.linux.lib.musl.general:7452 Archived-At: On Mon, Apr 20, 2015 at 10:21:31PM +0300, Alexander Monakov wrote: > After recent commits musl fails to build for x86_64 and x32. > > When configuring on x86_64, configure mistakenly thinks the host compiler is > going to produce code for x32. This is because __ILP32__ check wrongly > succeeds. That is because it's looking for compilation failure due to #error > preprocessor directive, but compilation instead fails because vis.h is not > found (-include vis.h is added to CFLAGS with -I giving the path to it). I > have used the following patch to proceed with the build: > > diff --git a/configure b/configure > index 0e39694..cebee09 100755 > --- a/configure > +++ b/configure > @@ -440,7 +440,7 @@ printf "%s\n" "$visibility" > fi > > if test "x$visibility" == xyes ; then > -CFLAGS_AUTO="$CFLAGS_AUTO -include vis.h" > +CFLAGS_AUTO="$CFLAGS_AUTO -I./arch/$ARCH -I src/internal -I./include -include vis.h" This is probably inappropriate. I can think of two reasonable alternatives; not sure which is better: 1. Have a global CFLAGS_PREDEF that that gets applied to all tests but not added to CFLAGS_AUTO. 2. Just use -include src/internal/vis.h. I kinda like #2 better. > CFLAGS_AUTO="${CFLAGS_AUTO# }" > fi > > > Would be nice to rework trycppif configure function to be robust against this > kind of failures. > > When targeting x32, build fails due to "jmp" instruction expecting a 64-bit > argument. At least the following change is needed, but I'm not sure it's > sufficient: the register should be zero-extended, not sign-extended: > > diff --git a/arch/x32/reloc.h b/arch/x32/reloc.h > index 7c72d26..492fbf1 100644 > --- a/arch/x32/reloc.h > +++ b/arch/x32/reloc.h > @@ -23,4 +23,4 @@ > #define REL_TPOFF R_X86_64_TPOFF64 > > #define CRTJMP(pc,sp) __asm__ __volatile__( \ > - "mov %1,%%esp ; jmp *%0" : : "r"(pc), "r"(sp) : "memory" ) > + "mov %1,%%esp ; jmp *%q0" : : "r"(pc), "r"(sp) : "memory" ) + "mov %1,%%esp ; jmp *%0" : : "r"((uint64_t)(uintptr_t)pc), "r"(sp) : "memory" ) Rich