[PATCH RFC] bfd/ELF: restrict file alignment for object files

[Jan Beulich]

On 09.09.2024 15:46, Nick Clifton wrote:
> Hi Jan,
> 
>>>> yet then - any hint from where to take the two alignment values? Do they
>>>> need adding from scratch, for every architecture?
>>>
>>> Maybe bfd->xvec->section_alignment_power ?
>>>
> 
>> I'm afraid I can't even spot a field of this name.
> 
> Which would be because I was referencing the wrong structure.  Doh.
> I meant the bfd_arch_info structure from bfd/archures.c:603...
> 
>    typedef struct bfd_arch_info
>    {
>      int bits_per_word;
>      int bits_per_address;
>      int bits_per_byte;
>      enum bfd_architecture arch;
>      unsigned long mach;
>      const char *arch_name;
>      const char *printable_name;
>      unsigned int section_align_power;
>    [...]
> 
>> If there was one, it
>> also couldn't possibly express both minimal code and minimal data _file_
>> alignment (we're not really talking of in-memory alignment here, after
>> all).
> 
> Hmm, since we are talking about sections rather than segments, I would
> think that we can assume that there would never be a need for a in-file
> alignment to be greater than an in-memory alignment.  Hence we can use
> whatever in-memory section alignment fields we may have as a guide for
> the maximum file alignment needed.
> 
> But for a minimum in-file works-for-all-sections alignment, I do not think
> that we have a structure field that we can use.  We can always add one of
> course.  Maybe just using a default value (eg 4) and allowing targets to
> override it if necessary would work ?

Just fyi that I think I found what I was after: struct elf_size_info's
log_file_align. That's used elsewhere for very similar purposes.

Jan