Hi,

The patch is fixed some problems as your comments in previous mails.

>The assumption that the uleb128 address size can only decrease with relaxation is true only if both addresses are in the same section.
I found this kind of subtract is banned in cvt_frag_to_fill for
rs_leb128.  Therefore, it doesn't have to restrict by ourselves.

Thanks for your reviews.

Nelson Chu <nelson.chu@sifive.com> 於 2019年12月18日 週三 上午9:50寫道：
>
> There is one minor thing.  Maciej has pointed out before that it's
> better done with one the percent-codes to `_bfd_error_handler' rather
> than aborting the link right away, so that any further link errors are
> also reported and you don't have to shake them out one by one.  So
> report the relocation error via linker's callback function seems to be
> better.  I think maybe we can report a dangerous relocation for the
> mismatched R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128.  What I meant
> is that,
>
> + case R_RISCV_SET_ULEB128:
> ...
> +  else
> +    {
> +      if (uleb128_rel->r_offset != rel->r_offset)
> + {
> -  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> - input_bfd, howto->name);
> -  bfd_set_error (bfd_error_bad_value);
> + msg = ("R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128 are mismatched. ");
> + r = bfd_reloc_dangerous;
> + break;
> + }
> +      relocation = relocation - uleb128_vma;
> +      uleb128_rel = NULL;
> +      break;
> +    }
> +
> + case R_RISCV_SUB_ULEB128:
> +  if (uleb128_rel)
> +    {
> +      if (uleb128_rel->r_offset != rel->r_offset)
> + {
> -  (*_bfd_error_handler) (_("%pB: relocation %s mismatched. "),
> - input_bfd, howto->name);
> -  bfd_set_error (bfd_error_bad_value);
> + msg = ("R_RISCV_SET_ULEB128 and R_RISCV_SUB_ULEB128 are mismatched. ");
> + r = bfd_reloc_dangerous;
> + break;
> + }
> +      relocation = uleb128_vma - relocation;
> +      uleb128_rel = NULL;
> +      break;
> +    }
>
> Thanks and regards
> Nelson



-- 
Best regards,
Kuan-Lin Chen.
kuanlinchentw@gmail.com