[llvm-dev] Building llvm with clang and lld on arm and the llvm arm backend relocation on position independent code

[Rui Ueyama via llvm-dev]

On Sun, Jun 25, 2017 at 6:40 AM, Alessandro Pistocchi via llvm-dev <
llvm-dev at lists.llvm.org> wrote:

> Hi,
>
> I am trying to build a completely GNU free linux toolchain for the
> raspberry pi.
>
> I successfully managed to compile llvm and clang for armv7 hard float (
> both as a cross compiler and as a native compiler ) together with the
> following:
>
> Llvm with clang and lld
> Clang builtins
> Musl libc
> libc++, libc++abi, libunwind
>
> All works well with the only thing to notice being the need to use -fPIC
> in order to access some library functions when my own c/c++ programs access
> those functions.
>
> The fact seems to be that musl libc exports some symbols as protected (
> probably correctly ) and lld ( probably correctly ) says it cannot preempt
> those symbols.
> For this reason I seem to have to use -fPIC in the C and CXX flags but
> everything seems to work ok.
>

Is this the same problem mentioned in
https://bugs.llvm.org/show_bug.cgi?id=32425?


> Then I tried to use this compiler ( both the cross compiler and the native
> compiler ) to compile llvm + clang + lld ( I want to have the toolchain
> built with itself, again without any GNU software involved ) but when
> building the clang executable I ran into the arm relocation problems
> mentioned here in the “Hacks” section when using -fPIC:
> http://llvm.org/docs/HowToCrossCompileLLVM.html .
> On the other hand, I seem to need -fPIC otherwise cmake fails to find some
> libc functions such as futimes/futimens and many others. If I use -fPIC for
> CFLAGS but not for CXXFLAGS then cmake finds those symbols but then
> obviously fails at a later stage with lld unable to preempt those symbols.
>
> This seems to be a conflict I cannot solve without someone within the llvm
> team fixing the arm relocation problem. Is there any estimate of when this
> could happen? I am happy to spend the time testing any solutions by
> building my own toolchain until it succeeds. Or is there any other solution?
>
> Thank you,
> Alex
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170626/d45de33e/attachment.html>