Re: [Qemu-devel] [ARM/FDPIC v2 2/4] linux-user: ARM-FDPIC: Identify ARM FDPIC binaries

[Christophe Lyon]

On 23/04/2018 14:17, Peter Maydell wrote:
> On 23 April 2018 at 08:51, Christophe Lyon <address@hidden> wrote:
> > Define an ARM-specific version of elf_is_fdpic:
> > FDPIC ELF objects are identified with e_ident[EI_OSABI] ==
> > ELFOSABI_ARM_FDPIC.
> >
> > Co-Authored-By: Mickaël Guêné <address@hidden>
> > Signed-off-by: Christophe Lyon <address@hidden>
> >
> > diff --git a/include/elf.h b/include/elf.h
> > index c0dc9bb..934dbbd 100644
> > --- a/include/elf.h
> > +++ b/include/elf.h
> > @@ -1483,6 +1483,7 @@ typedef struct elf64_shdr {
> >   #define ELFOSABI_TRU64          10      /* Compaq TRU64 UNIX.  */
> >   #define ELFOSABI_MODESTO        11      /* Novell Modesto.  */
> >   #define ELFOSABI_OPENBSD        12      /* OpenBSD.  */
> > +#define ELFOSABI_ARM_FDPIC      65      /* ARM FDPIC */
> >   #define ELFOSABI_ARM            97      /* ARM */
> >   #define ELFOSABI_STANDALONE     255     /* Standalone (embedded) application 
> > */
> >
> > diff --git a/linux-user/elfload.c b/linux-user/elfload.c
> > index bbe93b0..76d7718 100644
> > --- a/linux-user/elfload.c
> > +++ b/linux-user/elfload.c
> > @@ -1681,11 +1681,18 @@ static void zero_bss(abi_ulong elf_bss, abi_ulong 
> > last_bss, int prot)
> >       }
> >   }
> >
> > +#ifdef TARGET_ARM
> > +static int elf_is_fdpic(struct elfhdr *exec)
> > +{
> > +    return exec->e_ident[EI_OSABI] == ELFOSABI_ARM_FDPIC;
> > +}
> > +#else
> >   /* Default implementation, always false.  */
> >   static int elf_is_fdpic(struct elfhdr *exec)
> >   {
> >       return 0;
> >   }
> > +#endif
>
> I have a strong dislike for per-target ifdef ladders. Can we instead
> put the target's implementation of elf_is_fdpic() into
> linux-user/$ARCH/target_elf.h
> and also have that header do
> #define TARGET_HAS_ELF_FDPIC
>
> and then in the generic code we can protect the default elf_is_fdpic()
> with #ifndef TARGET_HAS_ELF_FDPIC.

How invasive could that be?
Your proposal is appealing, but target_elf.h is only included by 
linux-user/main.c, which does not define elfhdr etc...
All that knowledge is in linux-user/elfload.c, which controls what 
include/elf.h defines.

Should I re-engineer that?

Thanks,

Christophe


> thanks
> -- PMM
> .