libffi build failure with ld.gold (works with trunk 20110106, fails with 20110110)

[Matthias Klose]

Binary package hint: gcc-snapshot

this is a build enabled to build Go, and pulling in libffi, which fails with

libtool: link: /scratch/packages/gcc/4.6/gcc-4.6-4.6-20110110/build/./gcc/xgcc
-B/scratch/packages/gcc/4.6/gcc-4.6-4.6-20110110/build/./gc
c/ -B/usr/i686-linux-gnu/bin/ -B/usr/i686-linux-gnu/lib/ -isystem
/usr/i686-linux-gnu/include -isystem /usr/i686-linux-gnu/sys-include
-shared src/.libs/debug.o src/.libs/prep_cif.o src/.libs/types.o
src/.libs/raw_api.o src/.libs/java_raw_api.o src/.libs/closures.o
src/x86/.libs/ffi.o src/x86/.libs/sysv.o -Wl,-soname -Wl,libffi.so.4 -o
.libs/libffi.so.4.0.1
ld.gold: internal error in set_offset, at ../../gold/output.cc:4030
collect2: ld returned 1 exit status
make[6]: *** [libffi.la] Error 1

I didn't change the build environment between the two builds. gcc was
configured with

         --enable-languages=c,c++,fortran,objc,obj-c++,go
         --prefix=/usr
         --program-suffix=-4.6
         --enable-shared
         --with-system-zlib
         --without-included-gettext
         --enable-threads=posix
         --enable-nls
         --enable-clocale=gnu
         --enable-libstdcxx-debug
         --enable-libstdcxx-time=yes
         --enable-plugin
         --enable-gold
         --enable-ld=default
         --with-plugin-ld=ld.gold
         --enable-objc-gc
         --enable-targets=all
         --with-arch-32=i686
         --with-tune=generic
         --enable-checking=release
         --build=i686-linux-gnu
         --host=i686-linux-gnu
         --target=i686-linux-gnu

[Matthias Klose (doko)]

Binary package hint: gcc-snapshot

this is a build enabled to build Go, and pulling in libffi, which fails with

libtool: link: /scratch/packages/gcc/4.6/gcc-4.6-4.6-20110110/build/./gcc/xgcc
-B/scratch/packages/gcc/4.6/gcc-4.6-4.6-20110110/build/./gc
c/ -B/usr/i686-linux-gnu/bin/ -B/usr/i686-linux-gnu/lib/ -isystem
/usr/i686-linux-gnu/include -isystem /usr/i686-linux-gnu/sys-include
-shared src/.libs/debug.o src/.libs/prep_cif.o src/.libs/types.o
src/.libs/raw_api.o src/.libs/java_raw_api.o src/.libs/closures.o
src/x86/.libs/ffi.o src/x86/.libs/sysv.o -Wl,-soname -Wl,libffi.so.4 -o
.libs/libffi.so.4.0.1
ld.gold: internal error in set_offset, at ../../gold/output.cc:4030
collect2: ld returned 1 exit status
make[6]: *** [libffi.la] Error 1

I didn't change the build environment between the two builds. gcc was
configured with

         --enable-languages=c,c++,fortran,objc,obj-c++,go
         --prefix=/usr
         --program-suffix=-4.6
         --enable-shared
         --with-system-zlib
         --without-included-gettext
         --enable-threads=posix
         --enable-nls
         --enable-clocale=gnu
         --enable-libstdcxx-debug
         --enable-libstdcxx-time=yes
         --enable-plugin
         --enable-gold
         --enable-ld=default
         --with-plugin-ld=ld.gold
         --enable-objc-gc
         --enable-targets=all
         --with-arch-32=i686
         --with-tune=generic
         --enable-checking=release
         --build=i686-linux-gnu
         --host=i686-linux-gnu
         --target=i686-linux-gnu

[Matthias Klose (doko)]

ld.gold from binutils 2.21 was used

[Matthias Klose (doko)]

ld.gold from binutils 2.21 was used.

object files at
https://bugs.launchpad.net/gcc/+bug/701110/+attachment/1790281/+files/tst.tar.xz

[Ian Lance Taylor (ianlancetaylor)]

This problem is occurring because the file sysv.o has a writable .eh_frame section. That is confusing gold's relro support. This is certainly a bug in gold. However, there does also seem to be a problem in libffi. Why is the libffi configure.ac setting EH_FRAME_FLAGS to "aw"?

As far as I can see, the test in libffi/configure.ac is bogus. It is compiling code and looking for the setting of the .eh_frame section. However, the code it is compiling has no need to handle exceptions and therefore has no .eh_frame section. The configure script draws the wrong conclusion and emits a writable .eh_frame section when it should emit a readable one. This should be fixed independently of the gold problem.

[Ian Lance Taylor (ianlancetaylor)]

I've committed the fix to gold to both binutils mainline and the binutils 2.21 branch.

http://sourceware.org/ml/binutils/2011-01/msg00142.html

[Andreas Schwab (schwab-linux-m68k)]

The assembler files should really be using cfi directives like m68k/sysv.S.

[Matthias Klose (doko)]

closing obsolete issue

[Matthias Klose (doko)]

closing ancient issue