Re: [PATCH] ARM: use ARM unwinder for gcov

[Arnd Bergmann]

On Tue, May 5, 2020 at 4:19 PM Russell King - ARM Linux admin
<linux@xxxxxxxxxxxxxxx> wrote:
>
> On Tue, May 05, 2020 at 04:11:56PM +0200, Arnd Bergmann wrote:
> > Using gcov on ARM fails when the frame pointer unwinder is used:
> >
> > arm-linux-gnueabi-ld: kernel/softirq.o:(.ARM.exidx+0x120): undefined reference to `__aeabi_unwind_cpp_pr0'
> > arm-linux-gnueabi-ld: init/main.o:(.ARM.exidx+0x98): undefined reference to `__aeabi_unwind_cpp_pr0'
> > arm-linux-gnueabi-ld: init/version.o:(.ARM.exidx+0x0): undefined reference to `__aeabi_unwind_cpp_pr0'
> > arm-linux-gnueabi-ld: init/do_mounts.o:(.ARM.exidx+0x28): undefined reference to `__aeabi_unwind_cpp_pr0'
> > arm-linux-gnueabi-ld: init/do_mounts_initrd.o:(.ARM.exidx+0x0): undefined reference to `__aeabi_unwind_cpp_pr0'
> > arm-linux-gnueabi-ld: init/initramfs.o:(.ARM.exidx+0x8): more undefined references to `__aeabi_unwind_cpp_pr0' follow
> >
> > This is likely a bug in clang that should be fixed in the compiler.
> > Forcing the use of the ARM unwinder in this configuration however
> > works around the problem.
>
> Or should the stub functions in arch/arm/kernel/unwind.c be moved out?

I had not realized this is only a stub. Moving them out should fix the
link failure, I can
send a patch after testing if we are sure that calling them is not a
bug in clang
that would lead to runtime failures.

       Arnd