From patchwork Fri Mar 1 16:41:59 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Florian Weimer X-Patchwork-Id: 31697 Received: (qmail 60995 invoked by alias); 1 Mar 2019 16:42:08 -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 60947 invoked by uid 89); 1 Mar 2019 16:42:07 -0000 Authentication-Results: sourceware.org; auth=none X-Spam-SWARE-Status: No, score=-26.9 required=5.0 tests=BAYES_00, GIT_PATCH_0, GIT_PATCH_1, GIT_PATCH_2, GIT_PATCH_3, SPF_HELO_PASS autolearn=ham version=3.3.2 spammy= X-HELO: mx1.redhat.com From: Florian Weimer To: Paul Pluzhnikov Cc: GLIBC Devel , "Carlos O'Donell" Subject: Re: [PATCH] elf/tst-big-note: Improve accuracy of test [BZ #20419] References: <878sxyy8id.fsf@oldenburg2.str.redhat.com> Date: Fri, 01 Mar 2019 17:41:59 +0100 In-Reply-To: (Paul Pluzhnikov's message of "Fri, 1 Mar 2019 08:18:12 -0800") Message-ID: <87a7iev960.fsf@oldenburg2.str.redhat.com> User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/26.1 (gnu/linux) MIME-Version: 1.0 * Paul Pluzhnikov: > On Fri, Mar 1, 2019 at 6:28 AM Florian Weimer wrote: >> >> It is possible that the link editor injects an allocated ABI tag note >> before the artificial, allocated large note in the test. > > Looks good to me. Thanks! > >> +$(objpfx)tst-big-note-lib.so: $(objpfx)tst-big-note-lib.o >> + $(LINK.o) -shared -o $@ $(LDFLAGS.so) $< > > I got lost tracing through Makefile machinery trying to find how > -nostartfiles gets into the link line here, but I'll take your word > for it :-) I get this command (slightly gcc -m32 -nostdlib -nostartfiles -o /BUILD/elf/tst-big-note -Wl,-z,combreloc -Wl,-z,relro -Wl,--hash-style=both /BUILD/csu/crt1.o /BUILD/csu/crti.o `gcc -m32 --print-file-name=crtbegin.o` /BUILD/elf/tst-big-note.o /BUILD/support/libsupport_nonshared.a /BUILD/elf/tst-big-note-lib.so -Wl,-dynamic-linker=/lib/ld-linux.so.2 -Wl,-rpath-link=/BUILD:/BUILD/math:/BUILD/elf:/BUILD/dlfcn:/BUILD/nss:/BUILD/nis:/BUILD/rt:/BUILD/resolv:/BUILD/mathvec:/BUILD/support:/BUILD/crypt:/BUILD/nptl /BUILD/libc.so.6 /BUILD/libc_nonshared.a -Wl,--as-needed /BUILD/elf/ld.so -Wl,--no-as-needed -lgcc -Wl,--as-needed -lgcc_s -Wl,--no-as-needed `gcc -m32 --print-file-name=crtend.o` /BUILD/csu/crtn.o But looking at the build output, I saw new warnings: Makefile:1522: warning: overriding recipe for target '/home/fweimer/src/gnu/glib c/build/elf/tst-big-note-lib.so' ../Makerules:769: warning: ignoring old recipe for target '/home/fweimer/src/gnu/glibc/build/elf/tst-big-note-lib.so' I missed one further adjustment in the filtmod1 case I needed to mirror, but with the new patch, the warning is gone. Thanks, Florian elf/tst-big-note: Improve accuracy of test [BZ #20419] It is possible that the link editor injects an allocated ABI tag note before the artificial, allocated large note in the test. Note parsing in open_verify stops when the first ABI tag note is encountered, so if the ABI tag note comes first, the problematic code is not actually exercised. Also tweak the artificial note so that it is a syntactically valid 4-byte aligned note, in case the link editor tries to parse notes and process them. Improves the testing part of commit 0065aaaaae51cd60210ec3a7e13. 2019-03-01 Florian Weimer [BZ #20419] * elf/tst-big-note-lib.S: Create a syntactically valid note. * elf/Makefile (tst-big-note-lib.so): Do not link with startup code, to avoid creating an ABI tag note. (modules-names-nobuild): Add tst-big-note-lib. Reviewed-by: Carlos O'Donell diff --git a/elf/Makefile b/elf/Makefile index 55204073a3..310a37cc13 100644 --- a/elf/Makefile +++ b/elf/Makefile @@ -316,8 +316,8 @@ endif modules-execstack-yes = tst-execstack-mod extra-test-objs += $(addsuffix .os,$(strip $(modules-names))) -# filtmod1.so has a special rule -modules-names-nobuild := filtmod1 +# filtmod1.so, tst-big-note-lib.so have special rules. +modules-names-nobuild := filtmod1 tst-big-note-lib tests += $(tests-static) @@ -1515,6 +1515,11 @@ tst-libc_dlvsym-static-ENV = \ $(objpfx)tst-libc_dlvsym-static.out: $(objpfx)tst-libc_dlvsym-dso.so $(objpfx)tst-big-note: $(objpfx)tst-big-note-lib.so +# Avoid creating an ABI tag note, which may come before the +# artificial, large note in tst-big-note-lib.o and invalidate the +# test. +$(objpfx)tst-big-note-lib.so: $(objpfx)tst-big-note-lib.o + $(LINK.o) -shared -o $@ $(LDFLAGS.so) $< $(objpfx)tst-unwind-ctor: $(objpfx)tst-unwind-ctor-lib.so diff --git a/elf/tst-big-note-lib.S b/elf/tst-big-note-lib.S index e2008cf4ae..721686fa0e 100644 --- a/elf/tst-big-note-lib.S +++ b/elf/tst-big-note-lib.S @@ -20,7 +20,13 @@ On a typical Linux system with 8MiB "ulimit -s", that was enough to trigger stack overflow in open_verify. */ +#define NOTE_SIZE 8*1024*1024 + .pushsection .note.big,"a" -.balign 4 -.fill 8*1024*1024, 1, 0 + .balign 4 + .long 5 /* n_namesz. Length of "GLIBC". */ + .long NOTE_SIZE /* n_descsz. */ + .long 0 /* n_type. */ + .ascii "GLIBC\0\0\0" /* Name and alignment to four bytes. */ + .fill NOTE_SIZE, 1, 0 .popsection