This is the mail archive of the binutils@sourceware.org mailing list for the binutils project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

-Ttext and phdr_in_segment


Hi,

 _bfd_elf_map_sections_to_segments sets phdr_in_segment to
 TRUE and sets it to false only if the LMA of the first section is less
 than the elf header size + program headers size. 

      bfd_size_type phdr_size = elf_tdata (abfd)->program_header_size;

      if (phdr_size == (bfd_size_type) -1)
        phdr_size = get_program_header_size (abfd, info);                                                                                                                          
      phdr_size += bed->s->sizeof_ehdr;
      if ((abfd->flags & D_PAGED) == 0
          || (sections[0]->lma & addr_mask) < phdr_size
          <snip>)
        phdr_in_segment = FALSE;
    }

 Is the comparison with LMA right for embedded targets like AVR, where
 the LMA refers to a FLASH address, and the header is not written
 to FLASH? What is the intention behind the check?

 For the AVR target, specifying -Ttext=<val>, with <val> bigger than 
 the size of the headers results in phdr_in_segment being set to true,
 and this causes the headers to be placed in the text segment. 

 For <val> less than sizeof(headers), say 0x40, this is what objdump prints
 Program Header:
    LOAD off    0x00000074 vaddr 0x00000040 paddr 0x00000040 align 2**0
         filesz 0x00000112 memsz 0x00000112 flags r-x

 With a bigger <val> (-Ttext=x100), this is the output of objdump.

 Program Header:
    LOAD off    0x00000000 vaddr 0x0000008c paddr 0x0000008c align 2**0
         filesz 0x00000186 memsz 0x00000186 flags r-x
   <snip>
 
 Notice that the file offset is zero, and the addr fields are set to 
 <val> - sizeof(headers) (0x100 - 0x74, in this case.) This confuses the
 "programmer", which reads the header information to write to FLASH - it
 ends up writing the ELF + program headers also, instead of just the
 .text section.

 My question, again, is if comparing the headers size against the LMA of
 the text section (section[0]) correct? Is there a way to "turn off"
 placement of the headers in a segment, rather than the start of the
 file?

Regards
Senthil


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]