Re: [PATCH v4 06/17] RISC-V: Startup and Dynamic Loading Code

[Palmer Dabbelt <palmer at dabbelt dot com>]

On Mon, 15 Jan 2018 08:54:55 PST (-0800), joseph@codesourcery.com wrote:
> On Sat, 13 Jan 2018, Palmer Dabbelt wrote:
>
> > diff --git a/sysdeps/riscv/dl-machine.h b/sysdeps/riscv/dl-machine.h
>
> > +/* Return nonzero iff ELF header is compatible with the running host.  */
> > +static inline int __attribute_used__
> > +elf_machine_matches_host (const ElfW (Ehdr) *ehdr)
> > +{
> > +  return ehdr->e_machine == EM_RISCV;
> > +}
>
> I don't think this is sufficient.  You should also check that the
> floating-point ABI matches so you don't load shared libraries for the
> wrong ABI.  (See e.g. MIPS for a more complicated elf_machine_matches_host
> example.)  If in future you have RV64I ILP32, you'll also need to check
> that only the right ILP32 libraries for the current instruction set are
> loaded (not RV32I into an RV64I ILP32 process or vice versa).  (Generic
> code in dl-load.c deals with only loading whichever is appropriate of
> 32-bit or 64-bit ELF.)

Ah, OK -- when I was going through something else I was wondering how that got 
enforced, I guess the answer is that it wasn't.  I think something like this 
should do it

   diff --git a/sysdeps/riscv/dl-machine.h b/sysdeps/riscv/dl-machine.h
   index 7d4fcee79662..21e401dbf2fb 100644
   --- a/sysdeps/riscv/dl-machine.h
   +++ b/sysdeps/riscv/dl-machine.h
   @@ -57,7 +57,20 @@
    static inline int __attribute_used__
    elf_machine_matches_host (const ElfW (Ehdr) *ehdr)
    {
   -  return ehdr->e_machine == EM_RISCV;
   +  /* We can only run RISC-V binaries.  */
   +  if (ehdr->e_machine != EM_RISCV)
   +    return 1;
   +
   +#ifdef __ricsv_float_abi_double
   +  /* Hard-float libraries */
   +  if ((ehdr->e_flags & EF_RISCV_FLOAT_ABI) == EF_RISCV_FLOAT_ABI_DOUBLE)
   +    return 1;
   +#else
   +  if ((ehdr->e_flags & EF_RISCV_FLOAT_ABI) == EF_RISCV_FLOAT_ABI_SOFT)
   +    return 1;
   +#endif
   +
   +  return 0;
    }
   
    /* Return the link-time address of _DYNAMIC.  */

I'll include something like that in our v5.