From patchwork Tue Jun 2 23:56:38 2015 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Roland McGrath X-Patchwork-Id: 479735 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from sourceware.org (server1.sourceware.org [209.132.180.131]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 7313C14027F for ; Wed, 3 Jun 2015 09:56:51 +1000 (AEST) Authentication-Results: ozlabs.org; dkim=pass (1024-bit key; unprotected) header.d=sourceware.org header.i=@sourceware.org header.b=RVPTzFed; dkim-atps=neutral DomainKey-Signature: a=rsa-sha1; c=nofws; d=sourceware.org; h=list-id :list-unsubscribe:list-subscribe:list-archive:list-post :list-help:sender:mime-version:content-type :content-transfer-encoding:from:to:subject:message-id:date; q= dns; s=default; b=BgH937j+mJWm9mOd33joWghLP5Dr/XCcuolsQx0BzNedif tzBWayVgomWyF6UCweHnMnQoktzHRxSEP68wlwblb5J70sjBXE7nghhC9ryk+8Xm 8kErf/8vCxbpIyynVCc/gkiCKeeoLmOnXEqA1Bu9/7wJAtIVxbK6LxfvOkLLY= DKIM-Signature: v=1; a=rsa-sha1; c=relaxed; d=sourceware.org; h=list-id :list-unsubscribe:list-subscribe:list-archive:list-post :list-help:sender:mime-version:content-type :content-transfer-encoding:from:to:subject:message-id:date; s= default; bh=NaesanJcSXtRqG7fF7mAv8VzEVM=; b=RVPTzFedyG4cOWIzaqYa llOfYhV1qAvHuTQJHDLCuBpU74ed9MVQXZmQUKwNpM7IUvHdC4r8T/vOjz+BPYY0 GzPyxxOr5Rio8w1RlBFqtXzWTO2HwCAOI3X4nbjhXJG5wcLOFaD2sK4H51cC6okw axmiC+lr3/axXa5BqJOgJxs= Received: (qmail 130200 invoked by alias); 2 Jun 2015 23:56:43 -0000 Mailing-List: contact libc-alpha-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: libc-alpha-owner@sourceware.org Delivered-To: mailing list libc-alpha@sourceware.org Received: (qmail 130186 invoked by uid 89); 2 Jun 2015 23:56:42 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-1.0 required=5.0 tests=AWL, BAYES_05, KAM_LAZY_DOMAIN_SECURITY autolearn=no version=3.3.2 X-HELO: topped-with-meat.com MIME-Version: 1.0 From: Roland McGrath To: "GNU C. Library" Subject: [COMMITTED PATCH] BZ#18383: Another test case, with TLS refs and defs in separate TUs. Message-Id: <20150602235638.52B7A2C3AB9@topped-with-meat.com> Date: Tue, 2 Jun 2015 16:56:38 -0700 (PDT) X-CMAE-Score: 0 X-CMAE-Analysis: v=2.1 cv=SvUDtp+0 c=1 sm=1 tr=0 a=WkljmVdYkabdwxfqvArNOQ==:117 a=14OXPxybAAAA:8 a=kj9zAlcOel0A:10 a=hOe2yjtxAAAA:8 a=mDV3o1hIAAAA:8 a=AAIEBi3OztIJHBSxz5sA:9 a=CjuIK1q_8ugA:10 In investigating BZ#18383 (the elf/tst-tlsalign case) on ARM, I found that it in fact appears to be an assembler bug. That bug is only tickled by the way GCC emits the references to the TLS variables when they are in the same translation unit. So this new test case (tst-tlsalign-extern) puts them in a separate file. On ARM, tst-tlsalign-extern works while tst-tlsalign fails. I think this demonstrates that libc is not actually buggy on ARM. However, the original case still fails when statically linked on i?86 and x86_64 (it crashes in early startup, which was never the ARM failure mode)--and so does tst-tlsalign-extern-static. So I think x86 has a bug that is distinct from the ARM bug, though one of the same tests tickles both bugs. Since my current focus is on ARM, I'm going to report the ARM assembler bug and not worry about this further for now. It would be worthwhile for someone to investigate the x86 bug, which is probably an actual libc bug. (It affects only static linking and only an unusual usage pattern, so it's not a high priority.) Thanks, Roland 2015-06-02 Roland McGrath [BZ #18383] * elf/tst-tlsalign-extern.c: New file. * elf/tst-tlsalign-extern-static.c: New file. * elf/tst-tlsalign-vars.c: New file. * elf/Makefile (tests-static): Add tst-tlsalign-extern-static. [$(build-shared) = yes] (tests): Add tst-tlsalign-extern. ($(objpfx)tst-tlsalign-extern): Depend on tst-tlsalign-vars.o. ($(objpfx)tst-tlsalign-extern-static): Likewise. (test-xfail-tst-tlsalign-extern-static): New variable. diff --git a/elf/Makefile b/elf/Makefile index dedf3c7..f2a115a 100644 --- a/elf/Makefile +++ b/elf/Makefile @@ -121,7 +121,8 @@ tests = tst-tls1 tst-tls2 tst-tls9 tst-leaks1 \ tst-auxv tests-static = tst-tls1-static tst-tls2-static tst-stackguard1-static \ tst-leaks1-static tst-array1-static tst-array5-static \ - tst-ptrguard1-static tst-dl-iter-static tst-tlsalign-static + tst-ptrguard1-static tst-dl-iter-static \ + tst-tlsalign-static tst-tlsalign-extern-static ifeq (yes,$(build-shared)) tests-static += tst-tls9-static tst-tls9-static-ENV = \ @@ -146,7 +147,7 @@ tests += loadtest restest1 preloadtest loadfail multiload origtest resolvfail \ tst-stackguard1 tst-addr1 tst-thrlock \ tst-unique1 tst-unique2 $(if $(CXX),tst-unique3 tst-unique4) \ tst-initorder tst-initorder2 tst-relsort1 tst-null-argv \ - tst-ptrguard1 tst-tlsalign + tst-ptrguard1 tst-tlsalign tst-tlsalign-extern # reldep9 ifeq ($(build-hardcoded-path-in-tests),yes) tests += tst-dlopen-aout @@ -528,10 +529,16 @@ $(objpfx)tst-initorder: $(objpfx)tst-initordera4.so $(objpfx)tst-initordera1.so $(objpfx)tst-null-argv: $(objpfx)tst-null-argv-lib.so $(objpfx)tst-tlsalign: $(objpfx)tst-tlsalign-lib.so -# BZ#18383: broken on at least ARM (both) and x86-64 (static only). +# BZ#18383: broken on at least ARM (both) and i386/x86-64 (static only). test-xfail-tst-tlsalign = yes test-xfail-tst-tlsalign-static = yes +$(objpfx)tst-tlsalign-extern: $(objpfx)tst-tlsalign-vars.o +$(objpfx)tst-tlsalign-extern-static: $(objpfx)tst-tlsalign-vars.o + +# BZ#18383: broken on at least i386/x86-64 (static only). +test-xfail-tst-tlsalign-extern-static = yes + tst-null-argv-ENV = LD_DEBUG=all LD_DEBUG_OUTPUT=$(objpfx)tst-null-argv.debug.out LDFLAGS-nodel2mod3.so = $(no-as-needed) LDFLAGS-reldepmod5.so = $(no-as-needed) diff --git a/elf/tst-tlsalign-extern-static.c b/elf/tst-tlsalign-extern-static.c new file mode 100644 index 0000000..e84900e --- /dev/null +++ b/elf/tst-tlsalign-extern-static.c @@ -0,0 +1 @@ +#include "tst-tlsalign-extern.c" diff --git a/elf/tst-tlsalign-extern.c b/elf/tst-tlsalign-extern.c new file mode 100644 index 0000000..661b671 --- /dev/null +++ b/elf/tst-tlsalign-extern.c @@ -0,0 +1,74 @@ +/* Test for large alignment in TLS blocks (extern case), BZ#18383. + Copyright (C) 2015 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + . */ + +#include +#include +#include + +/* This is the same as tst-tlsalign-static.c, except that it uses + TLS variables that are defined in a separate translation unit + (ts-tlsalign-vars.c). It turned out that the cause of BZ#18383 + on ARM was actually an ARM assembler bug triggered by the ways of + using .tdata/.tbss sections and relocs referring to them that GCC + chooses when the variables are defined in the same translation + unit that contains the references. */ + +extern __thread int tdata1; +extern __thread int tdata2; +extern __thread int tdata3; +extern __thread int tbss1; +extern __thread int tbss2; +extern __thread int tbss3; + +static int +test_one (const char *which, unsigned int alignment, int *var, int value) +{ + uintptr_t addr = (uintptr_t) var; + unsigned int misalign = addr & (alignment - 1); + + printf ("%s TLS address %p %% %u = %u\n", + which, (void *) var, alignment, misalign); + + int got = *var; + if (got != value) + { + printf ("%s value %d should be %d\n", which, got, value); + return 1; + } + + return misalign != 0; +} + +static int +do_test (void) +{ + int fail = 0; + + fail |= test_one ("tdata1", 4, &tdata1, 1); + fail |= test_one ("tdata2", 0x10, &tdata2, 2); + fail |= test_one ("tdata3", 0x1000, &tdata3, 4); + + fail |= test_one ("tbss1", 4, &tbss1, 0); + fail |= test_one ("tbss2", 0x10, &tbss2, 0); + fail |= test_one ("tbss3", 0x1000, &tbss3, 0); + + return fail ? EXIT_FAILURE : EXIT_SUCCESS; +} + +#define TEST_FUNCTION do_test () +#include "../test-skeleton.c" diff --git a/elf/tst-tlsalign-vars.c b/elf/tst-tlsalign-vars.c new file mode 100644 index 0000000..01b3501 --- /dev/null +++ b/elf/tst-tlsalign-vars.c @@ -0,0 +1,28 @@ +/* This is for tst-tlsalign-extern.c, which see. It's essential for the + purpose of the test that these definitions be in a separate translation + unit from the code using the variables. */ + +__thread int tdata1 = 1; +__thread int tdata2 __attribute__ ((aligned (0x10))) = 2; +__thread int tdata3 __attribute__ ((aligned (0x1000))) = 4; +__thread int tbss1; +__thread int tbss2 __attribute__ ((aligned (0x10))); +__thread int tbss3 __attribute__ ((aligned (0x1000))); + +/* This function is never called. But its presence in this translation + unit makes GCC emit the variables above in the order defined (perhaps + because it's the order in which they're used here?) rather than + reordering them into descending order of alignment requirement--and so + keeps it more similar to the tst-tlsalign-static.c case--just in case + that affects the bug (though there is no evidence that it does). */ + +void +unused (void) +{ + tdata1 = -1; + tdata2 = -2; + tdata3 = -3; + tbss1 = -4; + tbss2 = -5; + tbss3 = -6; +}