[U-Boot] [BUG] efi-x86_app_defconfig does not build with gcc 8.2

[Andy Shevchenko]

On Tue, Oct 09, 2018 at 07:28:36AM +0800, Bin Meng wrote:
> Hi Andy,
> 
> On Tue, Oct 9, 2018 at 12:31 AM Andy Shevchenko
> <andriy.shevchenko at linux.intel.com> wrote:
> >
> > On Mon, Oct 08, 2018 at 11:31:32PM +0800, Bin Meng wrote:
> > > +Andy
> > >
> > > On Sat, Sep 8, 2018 at 5:00 PM Heinrich Schuchardt <xypron.glpk at gmx.de> wrote:
> > > >
> > > > make mrproper && make efi-x86_app_defconfig && make
> > > >
> > > > results in
> > > >
> > > >   CC      common/image.o
> > > > {standard input}: Assembler messages:
> > > > {standard input}:21662: Error: junk at end of line, first unrecognized
> > > > character is `@'
> > > > {standard input}:21707: Error: junk at end of line, first unrecognized
> > > > character is `@'
> > > > {standard input}:21662: Error: can't resolve `end.6878'
> > > > {.u_boot_list_2_fit_loadable_3 section} - `start.6875'
> > > > {.u_boot_list_2_fit_loadable_1 section}
> > > > {standard input}:21707: Error: can't resolve `end.6878'
> > > > {.u_boot_list_2_fit_loadable_3 section} - `start.6875'
> > > > {.u_boot_list_2_fit_loadable_1 section}
> > > >
> > > > I am using Debian Buster with gcc-8.2.0.
> > > >
> > > > Building with gcc-7.3.0 works fine.
> > > >
> > > > We had a similar issue for qemu-x86_defconfig in
> > > > https://lists.denx.de/pipermail/u-boot/2018-July/336393.html
> > > > which has been resolved.
> > >
> > > I can reproduce this with kernel.org GCC 8.1.0 toolchain now. But I
> > > have no idea how this error is related to anything. Andy may have some
> > > idea.
> >
> > Does below fixes the issue (check also if the resulting binary runs)?
> >
> > --- a/arch/x86/config.mk
> > +++ b/arch/x86/config.mk
> > @@ -41,7 +41,7 @@ OBJCOPYFLAGS_EFI := -j .text -j .sdata -j .data -j .dynamic -j .dynsym \
> >  ifeq ($(IS_32BIT),y)
> >  CFLAGS_NON_EFI := -mregparm=3
> >  endif
> > -CFLAGS_EFI := -fpic -fshort-wchar
> > +CFLAGS_EFI := -fshort-wchar
> >
> 
> Thanks for checking. I tried this but the resulted binary does not
> boot. Hangs at "U-Boot EFI " and no console output anymore.

That's what I was suspecting after looking how EFI applications are being built.

So, someone familiar with compilers probably needs to check the following and act accordingly:
- either new compiler provides new switches to change behaviour as it was in GCC < 8, or
- create a small verifiable example with this linked list sections and submit a bug report to GCC, or
- U-Boot code (for linked lists) should be reconsidered from new requirements of the compiler, or
- EFI U-Boot application should have it's own relocation code (like normal U-Boot does), or
- ...anything I missed...

The issue from my limited knowledge looks like this:
EFI loader relies on the sections that are natural for dynamically linked
libraries (AKA .so), while U-Boot (when runs on bare metal) has another
approach, i.e. specific code that does a relocation based only on .reloc
information. The linked list sections in U-Boot are meant to be statically
compiled, which is not the case for EFI by some reason.

OTOH, it's clearly a compiler behaviour change, since with -fPIC -g it
generates DWARF that can't be translated by its own assembler!
(See option 2 in above list)

-- 
With Best Regards,
Andy Shevchenko