[mips] relative FDE relocations without DW_EH_PE_signed

[Mingjie Xing <mingjie dot xing at gmail dot com>]

Hello,

I get a test case failed when I run 'make check-ld' on my platform.
The failed test case is ld/testsuite/ld-mips-elf/eh-frame1.s. Here is
the error information.

regexp_diff match failure
regexp "^00000014 00000010 00000018 FDE cie=00000000 pc=00020000..00020010$"
line   "00000014 00000010 00000018 FDE cie=00000000 pc=100020000..100020010"
regexp_diff match failure
regexp "^00000028 00000010 0000002c FDE cie=00000000 pc=00020010..00020030$"
line   "00000028 00000010 0000002c FDE cie=00000000 pc=100020010..100020030"
regexp_diff match failure
...

FAIL: MIPS eh-frame 1, n32

I find that bfd will try to convert absolute FDE relocations into
relative ones by setting make_relative = 1. See bfd/elf-eh-frame.c
(_bfd_elf_parse_eh_frame):

          /* For shared libraries, try to get rid of as many RELATIVE relocs
             as possible.  */
          if (info->shared
              && (get_elf_backend_data (abfd)
                  ->elf_backend_can_make_relative_eh_frame
                  (abfd, info, sec)))
            {
              if ((cie->fde_encoding & 0xf0) == DW_EH_PE_absptr)
                this_inf->make_relative = 1;
              /* If the CIE doesn't already have an 'R' entry, it's fairly
                 easy to add one, provided that there's no aligned data
                 after the augmentation string.  */
              else if (cie->fde_encoding == DW_EH_PE_omit
                       && (cie->per_encoding & 0xf0) != DW_EH_PE_aligned)
                {
                  if (*cie->augmentation == 0)
                    this_inf->add_augmentation_size = 1;
                  this_inf->u.cie.add_fde_encoding = 1;
                  this_inf->make_relative = 1;
                }
            }

When bfd starts to write section in _bfd_elf_write_section_eh_frame,
the addresses will be converted as negative values, for example
0xfffeffe4. But the fde encoding is not DW_EH_PE_signed at this time.
This will cause the address be dumped as an unsigned value, which
looks like pc=100020000.

Is it a bug?

Thanks,
Mingjie