Re: [PATCH] (version 2) Change MIPS linker stubs to allow for more than 2^15 symbols.

[Richard Sandiford <richard at codesourcery dot com>]

Thiemo Seufer <ths@networkno.de> writes:
> Richard Sandiford wrote:
>> Thiemo Seufer <ths@networkno.de> writes:
>> > I felt the urge to do some more cleanups (interdiff appended),
>> > the only function change was to add the IRIX dummy entry to the
>> > worst case calculation.
>> 
>> Which IRIX dummy entry?  If you're referring to the dummy entry at
>> the beginning of the symbol table, that entry is a gABI feature and
>> is already accounted for.
>
> The one mentioned immediately below:
>
>   /* In the worst case, we'll get one stub per dynamic symbol, plus
>      to account for the dummy entry at the end required by IRIX
>      rld.  */
>
> and also
>
>   /* IRIX rld assumes that the function stub isn't at the end
>      of .text section.  So put a dummy.  XXX  */

I don't understand.  The code you changed in your patch is counting
_symbols_ not stubs:

  /* Get a worst-case estimate of the number of dynamic symbols needed.
     At this point, dynsymcount does not account for section symbols
     and count_section_dynsyms may overestimate the number that will
     be needed.  */
  dynsymcount = (elf_hash_table (info)->dynsymcount
		 + count_section_dynsyms (output_bfd, info));

That's the calculation you added 1 to.  The code that counts the number
of _stubs_ already takes the dummy stub into account.  You've quoted the
comment for that code yourself ;)

  /* In the worst case, we'll get one stub per dynamic symbol, plus
     one to account for the dummy entry at the end required by IRIX
     rld.  */
  loadable_size += htab->function_stub_size * (i + 1);

If we'd got that wrong, we'd have been missing the dummy stub before
the current patch series.  Note that here, "per dynaamic symbol" means
"per GOT-mapped dynamic symbol" (i.e. all those after DT_MIPS_GOTSYM).
"i" already holds that value, and its calculation is not affected
by the recent patches.

Richard