Re: cross-toolchain for x86 fails to compile

["H.J. Lu" <hjl.tools@gmail.com>]

On Mon, Feb 7, 2022 at 9:08 AM Adhemerval Zanella
<adhemerval.zanella@linaro.org> wrote:
>
>
>
> On 07/02/2022 12:16, Waldemar Brodkorb wrote:
> > Hi,
> > Adhemerval Zanella wrote,
> >
> >>
> >>
> >> On 04/02/2022 21:35, Waldemar Brodkorb wrote:
> >>> Hi,
> >>>
> >>> I am doing my regular update in OpenADK and updated to glibc 2.35.
> >>> All toolchains are build fine except x86.
> >>>
> >>> I am getting following error with binutils 2.37 and gcc 11.2:
> >>> /home/wbx/embedded-test/openadk/toolchain_qemu-x86_glibc/usr/bin/i686-openadk-linux-gnu-gcc
> >>> -o
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35-final/support/test-run-command
> >>> -nostdlib -nostartfiles -static -static-pie
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35-final/csu/rcrt1.o
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35-final/csu/crti.o
> >>> `/home/wbx/embedded-test/openadk/toolchain_qemu-x86_glibc/usr/bin/i686-openadk-linux-gnu-gcc
> >>> --print-file-name=crtbeginS.o`
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35-final/support/test-run-command.o
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35-final/elf/static-stubs.o
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35-final/support/libsupport_nonshared.a
> >>> -Wl,--start-group
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35-final/libc.a
> >>> -lgcc  -Wl,--end-group
> >>> `/home/wbx/embedded-test/openadk/toolchain_qemu-x86_glibc/usr/bin/i686-openadk-linux-gnu-gcc
> >>> --print-file-name=crtendS.o`
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35-final/csu/crtn.o
> >>> /home/wbx/embedded-test/openadk/toolchain_qemu-x86_glibc/usr/lib/gcc/i686-openadk-linux-gnu/11.2.0/../../../../i686-openadk-linux-gnu/bin/ld:
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35-final/libc.a(setlocale.o):
> >>> relocation R_386_GOT32 against absolute symbol
> >>> `_nl_current_LC_CTYPE_used' in section `__libc_freeres_fn' is
> >>> disallowed
> >>> collect2: error: ld returned 1 exit status
> >>> gmake[8]: *** [../Rules:293:
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35-final/support/test-run-command]
> >>> Error 1
> >>> gmake[7]: *** [Makefile:483: support/others] Error 2
> >>> gmake[6]: *** [Makefile:9: all] Error 2
> >>> gmake[5]: *** [Makefile:62:
> >>> /home/wbx/embedded-test/openadk/toolchain_build_qemu-x86_glibc/w-glibc-2.35-1/glibc-2.35/.compiled]
> >>> Error 2
> >>> gmake[4]: *** [Makefile:106: glibc-compile] Error 2
> >>> gmake[3]: *** [mk/build.mk:228: toolchain/final] Error 2
> >>> gmake[2]: *** [/home/wbx/embedded-test/openadk/mk/build.mk:177:
> >>> world] Error 2
> >>> gmake[1]: *** [Makefile.adk:25: all] Error 2
> >>> make: *** [Makefile:22: all] Error 2
> >>> build failed
> >>>
> >>> Is it a known bug in binutils or glibc?
> >>
> >> None we are aware of and we do build check on i686 constantly.  I just did a build
> >> with gcc 11.2 and binutins 2.37 and I haven't see any issue.  How did you build
> >> gcc and binutils?
> >
> > Thanks for your response. It seems TARGET CFLAGS are not allowed to
> > be -Os or any machine specific optimizations for x86. Forcing -O2
> > did work.
>
> I should work in fact, we have tracked multiple issues when building with -Os
> over the releases [1][2][3][4][5][6][7][8].
>
> The problem seems to the that now static-pie is enabled as default and it
> seems to be triggering this issue (with --disable-static-pie it does seem
> to work).
>
> The localedef.o is built with -fpie, as expected, so I am not sure if the
> issue is compiler that is emitting wrong relocations for -Os or binutils
> that is not expecting it.  With -O2 gcc emits R_386_GOT32X which seems to
> be fine.
>
> H.J or Siddhesh, could you check on this?
>
> [1] https://sourceware.org/bugzilla/show_bug.cgi?id=15105
> [2] https://sourceware.org/bugzilla/show_bug.cgi?id=19463
> [3] https://sourceware.org/bugzilla/show_bug.cgi?id=25240
> [4] https://sourceware.org/bugzilla/show_bug.cgi?id=22581
> [5] https://sourceware.org/bugzilla/show_bug.cgi?id=20729
> [6] https://sourceware.org/bugzilla/show_bug.cgi?id=19462
> [7] https://sourceware.org/bugzilla/show_bug.cgi?id=19465
> [8] https://sourceware.org/bugzilla/show_bug.cgi?id=19466

It looks like a linker bug:

https://sourceware.org/bugzilla/show_bug.cgi?id=28870

-- 
H.J.